debpalash

VoiceStudio 本地化语音生成工具

全离线语音克隆配音方案

22.2k 2.7k 90 1.2k 782 38

AI语音后端文本转语音语音转文本

(0 评价)

VoiceStudio 本地化语音生成工具

GitHub Trend · Star / Watch / Fork

使用 VoiceStudio，你可以：

开源本地语音工具，支持克隆、配音、转录，覆盖数百语种，替代在线语音生成服务。

用户评论 (0)

Github最新Release版本

v0.5.1

2026年08月28日

Changed

Model Catalogue now uses one breathable workspace canvas with simpler pane and engine-family navigation instead of nested cards and scroll regions (#1685)
Linux source launchers now catch missing libxdo and GStreamer audio plugins before they can cause a linker error or an aborted, blank WebKit renderer (#1680, #1682)
Dictation now carries one native output session from shortcut-down through final delivery, restores text, HTML, image, or file-list clipboards only when untouched, keeps Wayland copy-safe unless current-focus insertion is explicitly enabled, and retries silent Sherpa speech only through an already-installed local ASR model (#1175)
The backend binds its port immediately and reports startup progress live — /health answers 503-with-step and a new /startup/progress endpoint lists every step while PyTorch, API routes, and database migrations load in the background, so "starting at step X" is never mistakable for "dead"; the desktop splash narrates each step (#1550)

Added

A bundled Rust loopback sidecar exposes dictation start/stop/toggle, focused-output sessions, discovery, and JSON-RPC; the backend adds versioned streaming events and a dependency-free CLI bridge for Herdr, coding agents, editors, desktop apps, and TUIs (#1646)
Headless NVIDIA and ROCm machines can now join as worker-only Docker Compose services with no published UI and durable protocol-v2 enrollment; update both machines together before reconnecting (#1638) — thanks jkrogers9862!
Linux ARM64 (Asahi Apple Silicon) support for the OmniVoice GGUF engine — a linux-aarch64 binary built with GGML Vulkan where the toolchain allows it, so Apple GPUs accelerate generation through the open-source Honeykrisp driver instead of falling back to CPU-only (#1641)
One-command install on every desktop OS: curl -fsSL https://voicestudio.sh/install | sh (macOS/Linux/WSL) or irm https://voicestudio.sh/install | iex (Windows) — the URL serves the right script per platform, and Windows gains a source installer (scripts/install.ps1) with a 3-OS CI smoke (#1626)
Per-line subtitle management in the dub table: a line's end time is editable alongside its start (typing a time and dragging its timeline edge now take the same path), lines merge with the previous row as well as the next (Ctrl/Cmd+Shift+M), and a new line can be inserted into the gap after any row (#1612) — thanks invio-a11y!
CI now enforces performance regression budgets on the hot paths — operation-count tests pin streaming TTS to one synthesis per sentence and cached dub re-mixes to zero re-synthesis; fast-path guards cover zero re-decoding and ⌈N/W⌉ native batch calls when enabled (#1594)
Default-engine dubbing now synthesizes several segments per forward pass instead of one call per line — the width follows the host's device headroom (1 on CPU and low-VRAM cards, up to 8), OMNIVOICE_DUB_BATCH_WIDTH overrides it, and engines without native batching keep the single-segment path (#1594)
/ws/tts now reports real time-to-first-audio, and its RTF measures synthesis alone so a slow client can't inflate it (#1594)
The locally cached AudioSeal watermark generator warms on a background thread ~35s after boot (OMNIVOICE_PRELOAD_WATERMARK=0 opts out; explicitly setting =1 may download it), so the first synthesis no longer serializes the audioseal import + model load inline — measured at ~42s on a cold filesystem, 3s short of a 90s client timeout (#1576) — thanks paoloantinori!
Voices you've cloned stay "warm" across restarts — encoded references now persist to disk (~10 KB each), so the first generation of a session skips the re-encode and any transcription pass; OMNIVOICE_PROMPT_DISK_CACHE=0 opts out (#1565)
Optional FlashInfer acceleration for the default engine on CUDA (OMNIVOICE_FLASHINFER=1, ~2.2x measured) — needs the optional flashinfer-python package; missing package or kernel failure logs why and falls back to the standard path (#1565)
The bug reporter notices when you're on an outdated build and offers the latest release before filing — with a "File anyway" escape hatch — and stamps a Build status line into every report so up-to-date reports are tellable from stale ones (#1547)
Settings → Performance & Device gains a compute-device override (Auto / CUDA / ROCm / XPU / MPS / CPU, or OMNIVOICE_DEVICE) — pin the device when auto-detect picks wrong; only devices your machine actually has are offered (#1557)
Opt-in 24-layer PocketTTS checkpoints via OMNIVOICE_POCKETTTS_24L — better prosody for it/de/es/pt at roughly 2x render time (still faster than real-time); the fast 6-layer model stays the default (#1613) — thanks paoloantinori!

Docs

Supported-version and install guidance now identifies 0.5.1 as the stable desktop and container release (#1687)
The Docker Hub overview now shows the current engine-switching demo, Model Catalogue, and gallery voice workflow (#1593)
The Docker Hub overview and install guide now show the v0.5 tags and the built-in API-key/share-PIN security model instead of obsolete v0.4 and no-authentication guidance (#1592)
The READMEs now lead with download buttons and a three-step first-clone walkthrough, and a new benchmarks page anchors measured per-engine/per-device numbers on the in-repo harness (#1555)
Every engine now has its own guide — 21 new pages under docs/engines plus an index covering all 16 TTS and 11 ASR engines, linked from both READMEs (#1556)
The OmniVoice guide now covers combining style attributes with a reference clip (consistent instruct stabilizes cloning; the reference wins conflicts), inline pronunciation control (pinyin / CMU phonemes), and corrects the claim that the default engine can't do voice design — it can, from attributes (#1565)

Fixed

Workspaces now measure their responsive width when the post-bootstrap shell actually mounts, so native UI scaling reflows Projects and History instead of crushing the Dubbing demo into unreadable columns (#1683)
Dubbing keeps the source-language selector visible after a local file is chosen, so ASR can be pinned before transcription starts (#1678) — thanks Lonki-lomki-cloud!
First-run media-engine downloads become available to TTS immediately without a restart, and missing media-process failures now point to repair controls (#1677) — thanks farhataligpt-dev!
Source installs on AMD GPUs honour OMNIVOICE_TORCH_VARIANT=rocm: bun run desktop now swaps in the ROCm torch wheel after uv sync and launches the backend without re-syncing, instead of silently reverting to the CPU-only CUDA build on every start (#1665) — thanks uberclokr!
bun run desktop on a fresh clone no longer fails with "resource path ../../frontend/dist doesn't exist" — the dev launcher creates the placeholder Tauri resource directory before compiling (#1664) — thanks uberclokr!
macOS no longer loses TTS after the first request when Python lacks os.waitid; subprocess ownership now uses a safe waitpid fallback without risking reused process groups (#1656) — thanks paoloantinori!
Desktop startup, Retry, reset, uninstall, shutdown, and crash recovery now share one backend lifecycle owner; quitting interrupts first-run installers and gracefully drains then force-cleans the full backend process tree, so overlaps cannot duplicate or orphan it (#1635) — thanks Xohaibxobi!
Large Stories and Audiobook projects now persist in IndexedDB instead of overflowing the omnivoice.app localStorage envelope, with quota-safe migration and orderly exit/reload flushing (#1636) — thanks leodzai!
OmniVoice and its crash-isolated subprocess now route to AMD ROCm GPUs instead of warning and falling back to CPU (#1629) — thanks j4r3kb!
Dictation now cancels pending startup work, capture resources, sockets, and timers when the capture widget closes, preventing late work against a destroyed webview (#1645)
Streaming generation failures now show recognized recovery guidance and appear in Diagnostics instead of only returning a generic error (#1607)
The worker-capacity transport test no longer races its own setup: the 1-slot limit now goes through the enrollment handshake instead of mutating client config after connect, where the server's stream-open ConfigUpdate (carrying the registered capacity of 2) could overwrite it and fake an over-accept; failed CI twice on 2026-08-21 (#1630)
Moving words across a speaker boundary in a dub — merging two lines and splitting them again — no longer dubs the second half in the first speaker's voice; each half now keeps the speaker, voice, direction, gain, and language of whoever actually says it (#1612) — thanks invio-a11y!
Dictation on a WebView that refuses a 16 kHz audio context (WKWebView) now low-passes before downsampling, so frequencies above 8 kHz stop folding into the speech the recognizer is fed (#1610)
A microphone context that cannot be resumed now reports a mic error instead of leaving the dictation pill on "Listening" while capturing nothing (#1610)
Dictation no longer retains a whole session's audio for silent-model recovery — an open mic grew that buffer by ~115 MB an hour; the recent two minutes are kept instead (#1610)
The clipboard-delivery status is now translated in all 21 languages, so Wayland users — where clipboard delivery is the default — no longer see an English string (#1610)
A native sherpa-onnx load failure of any exception type now degrades to "engine unavailable" instead of taking the dictation WebSocket down (#1610)
Dictation now ships Whisper Tiny as its one cross-platform default, avoiding Parakeet's measured empty decoding on Windows while keeping Parakeet selectable behind runtime fallback (#1175)
Re-mixing a dub no longer decodes, rewrites, and re-reads every cached segment — same-rate cached audio is reused directly (and rejected if truncated), switching timing modes can't reuse slot-truncated audio as natural-rate, and RVC respects natural-rate modes (#1594)
PocketTTS French works again — pocket-tts only ships a 24-layer French model and rejected the name the sidecar asked for, so every French request failed at model load; French now always loads french_24l (#1613) — thanks paoloantinori!
Installing IndexTTS 2.5 no longer fails claiming an interrupted download — the weights repo ships config.yaml and VoiceStudio demanded a config_v2_5.yaml that exists in no upstream release; both names are accepted, so a hand-renamed checkout keeps working (#1611) — thanks zuiaiyutu!
IndexTTS 2.5 no longer has long-text generation killed at 60 seconds — the sidecar now proves it is alive every 5 seconds while infer() runs, and its deadline rises to 900s (OMNIVOICE_INDEXTTS_RECV_TIMEOUT_S) (#1611) — thanks zuiaiyutu!
The OpenAI-compatible /v1/audio/speech route now reuses the shared cached engine for explicit model ids instead of constructing a fresh engine — and its sidecar/model load, a ~28s floor per call for subprocess engines — on every request, with the same single-engine-resident discipline /generate applies (#1614) — thanks paoloantinori!
The setup wizard's RAM check no longer blocks 8 GB machines whose OS reports ~7.8 GB usable — the thresholds now tolerate reserved memory, and OMNIVOICE_RAM_PREFLIGHT=0 turns a genuine block into a warning for those who accept the OOM risk (#1618)
Invisible watermarking now runs eagerly instead of through torch.compile — AudioSeal's lazy compile sent the first embed of every session into Inductor's C++ codegen, which failed outright on macOS hosts whose toolchain couldn't serve it and shipped the audio unmarked after a 30-40s wait; first embed drops from 9.70s to 0.26s (#1615) — thanks paoloantinori!
The macOS Accessibility blocker now rechecks while visible and closes as soon as the grant is enabled instead of keeping a stale permission prompt on screen (#1609)
The dubbing editor's video and transcript columns can now be resized by pointer or keyboard, and the chosen split persists across launches (#1571) — thanks invio-a11y!
CPU-only synthesis now gets a bounded ten-minute execution budget, and a render that exhausts it is reported as a compute timeout instead of misleading "generation capacity is busy" queue pressure (#1588) — thanks ChienNguyen1111!
Rapid Launchpad ↔ Dub navigation now replaces the workspace DOM owner cleanly, so late media/waveform cleanup cannot trigger React's insertBefore crash (#1590) — thanks nicolas-jacques!
Watermark embedding failures now log the full traceback instead of just the exception message, so a silently-unmarked-audio incident (audio passes through unmarked by design) is diagnosable from the log alone (#1576) — thanks paoloantinori!
Dubbing now recovers rapid two-speaker exchanges when diarization collapses them, defaults new projects to lip sync without overwriting saved timing choices, and keeps the editor usable on narrow screens (#1584) — thanks victordonat0!
OMNIVOICE_ASR_BACKEND=omnivoice now selects the PyTorch-native Whisper path, so the documented ROCm escape hatch no longer fails as an unknown engine (#1582) — thanks patmansk!
Network Sharing from Windows MSI/portable installs now serves the bundled web interface to LAN devices instead of redirecting them to their own localhost (#1589) — thanks TWIISTED-STUDIOS!
Exported dubbed videos now mark the dubbed language as the default audio stream while keeping Original available as an explicit choice (#1575) — thanks invio-a11y!
Cloning references can no longer exhaust system memory: transcript-free clips up to 75 seconds are searched in five bounded passages, longer clips ask to be trimmed, and supplied transcripts remain capped at 20 seconds to preserve alignment (#1578) — thanks ACKAPOB!
Stored artifact subpaths now resolve after moving a data directory between Windows, macOS, Linux, and Docker, while traversal and symlink escapes remain blocked (#1559) — thanks Eman-Yousaf!
A remote browser hitting an API-key-configured server's admin 403 now gets the API-key login form instead of endless console 403s, while desktop and PIN-only/no-key servers keep the plain loopback error so guests are never offered a login no key can satisfy (#1568) — thanks paoloantinori!
The crash-isolated ASR sidecar and its download preflight now agree on which model to load — setting the shared faster-whisper model variable applies to both variants instead of the sidecar quietly using a different one (#1556)
"Ready" now requires the deep health probe (a working database-backed route), not just the identity probe — a backend whose install broke underneath can no longer be announced up while every real request fails (#1548)
Supervisor restarts after repeat crashes now back off (immediate, then 5s, then 15s) instead of respawning back-to-back, so a tight crash loop can't burn the whole restart budget in seconds (#1548)
The Linux desktop cleanup regression test now isolates build artifacts, so an existing developer build can no longer change its result (#1566)
Renaming, deleting, or revoking consent on a voice (and starring/clearing history, recording exports) now live-updates every open tab again — the sync routes' WebSocket events were silently dropped, which could look like "all my voices are gone" (#1561) — thanks paoloantinori!

CI

Project agents now share pinned Vite and FastAPI skills from skills.sh (#1594)
Weekly full-history secret scans no longer mistake the Ed25519 private-key type name for committed key material (#1591)

Linux x64 artifacts

43753aeb1f78e9318d8a4df97e45b76384cad5d88e6be423d696413f8329492c  VoiceStudio_0.5.1_amd64.AppImage
5bd13d39f8ccf8eb3ffedbefc27609ffa7a8cb90d24a8c5a421d19b4343d8578  VoiceStudio_0.5.1_amd64.AppImage.sig

macOS Intel artifacts

14a6d3dac76bd28ead71663a274d084d8d2d2a271301f4863a1691c45611236c  VoiceStudio_0.5.1_x64.dmg
9fb9e883cb2ee2bdde21d5d455d0e5334b6d2fb55541acdbd9aae130dc78876d  VoiceStudio.app.tar.gz
aed7f2ccd25b78d64dccc005e76b266ac0ebff03f97539b52b047fb9540af4fc  VoiceStudio.app.tar.gz.sig

Windows x64 artifacts

0e0a9b3ef7d4c48243bae2b12f610c5bc991bbb127e17700a410c24063cab2a6 *VoiceStudio_0.5.1_x64_en-US.msi
a3f7d688585b7532543accb8e98cc2fb1056f686bd4bc3358969a02f2f818cd3 *VoiceStudio_0.5.1_x64_en-US.msi.sig

Contributors

Thank you all 💜

详细ChangeLog

v0.5.0

2026年08月14日

Highlights

The app is now VoiceStudio (previously OmniVoice-Studio) — one waveform-and-spark identity across the app, docs and installers. Your data folder, settings and Docker image paths stay put.
Model Catalogue — engines and models in one workspace: every TTS, transcription and LLM engine with its device routing and install state, defaults picked there.
Switch TTS, ASR and LLM engines from the status bar or any workspace — ready-only choices, memory status, environment-pin protection, Ctrl/Cmd+E. (#1530)
Lend another machine's GPU with a join code and a QR scan — a Compute control in the status bar picks where jobs run, and several people can share one GPU box with revocable, certificate-pinned connections. (#1516, #1496)
Server mode is locked down: admin actions require an API key (#1525), and the remote UI exchanges it for short-lived sessions that never sit in browser storage or WebSocket URLs (#1528) — thanks bultodepapas!
A faster, cleaner Dub workspace for multilingual production, with a production command bar and per-language cards. (#1489)
The demo audio and video the app always advertised now actually ship, rendered by VoiceStudio's own engine. (#1517)
Dictation works on Wayland now — the portal shortcut actually fires (#1490, #1526) — and the recording pill is back on every desktop.
The Launchpad wears the project's signal-field waveform artwork over a quieter, borderless layout. (#1533)
The catalogue reads as headroom, not breakage: available engines sort first, uninstalled ones say what they need (#1531), and the LLM row names the provider that actually answers (#1538).
Gallery voices can be saved as local profiles — audio lands in your profile store with validated, content-addressed references. (#1542)

The Model Catalogue	The Voice Gallery

Changed

Gallery personas now preview through the local backend, retain their complete voice-design recipe, and open directly in Voice, Stories, or Audiobook. (#1542)
Support amount choices now use every theme's shared card, accent and focus tokens. (#1530)
Sponsoring, commercial licensing and getting in touch are one page now. They answered the same question between them and each used to live somewhere else, so they are three sections on a single scroll — the footer heart, the commercial-licence links and Contact all land on it, at the section you asked for. (#1522)
Model Catalogue switches panes with tabs instead of a two-state toggle, and the Engine Compatibility Matrix's TTS / ASR / LLM switcher is now tabs too — arrow-key navigable, and each tab still shows the engine it would use. (#1522)
Engines you can actually use sort to the top of the compatibility matrix, and an unavailable engine's name recedes instead of the whole row fading — the status badge and GPU chips that say why it is unavailable stay legible. (#1522)
Remote workers reads as a device list: status dot, address, latency, a live task meter, resident models and last-seen per machine, with housekeeping actions revealed on hover and a three-step empty state. (#1516)
The GPU picker and the new status-bar control paint their status dots and menu surfaces from themed tokens instead of fixed palette classes, so they stop showing Gruvbox colours on Midnight and Catppuccin. (#1516)
Dictation shows the pill again: a capture puts a small always-on-top capsule near the bottom of the screen you are working on — listening, transcribing, the result, and any error — and takes it away when the session ends. It never takes focus, so the text still lands in the app you were typing into. On Wayland the compositor decides where it sits; everywhere else it is bottom-centred.
Engines and models moved out of Settings into a new Model Catalogue workspace, reachable from the icon rail (or the title-bar tabs); Settings → Engines and Settings → Models now point there, and Settings keeps the models directory and Hugging Face mirror.
The Settings sidebar is keyboard-navigable: ⌘K / Ctrl+K jumps to the filter, ↑/↓ and Home/End move between categories, and Enter or ↓ from the filter drops into the list. Matching text in a filtered category name is highlighted, and group headers stay pinned while the list scrolls.
The Launchpad has a quieter, more spacious look: borderless feature tiles that light up on hover or keyboard focus, plain-numeral counts, hairline section rules, and one shared page column for the hero, tiles, recent files and project lists.
Linux release smoke now validates linuxdeploy's wrapped custom launcher instead of rejecting a healthy AppImage. (#1506)
Remote GPU workers render audiobooks chapter by chapter, with automatic per-chapter local fallback and one combined notice if the worker drops out. (#1478)
Remote GPU workers can now run a job to completion: long renders no longer die at two minutes, a worker that drops and reconnects mid-render keeps its work, and a timed-out job no longer takes the worker offline for good. Placing a job still needs the development-only POST /workers/tasks; wiring the app's own Synthesize button to it comes next.
Voice, Stories, Audiobook, Gallery, Settings, profiles, and Launchpad now use compact, responsive layouts with accessible controls. (#1491)
Dubbing's Generate Dub, Verify, and Export actions now use a compact hierarchy with visible labels, responsive reflow, and motion-safe feedback. (#1493)
The Dub workspace now has a compact production command bar, responsive flag-based language cards, media previews in Dub History, and a narrower Projects rail. (#1489)
VoiceStudio now uses one waveform-and-spark mark across the title bar, About screen, README, browser favicon, and every desktop/platform icon. (#1487)
PocketTTS now asks you to review its code license, model license and gated-access conditions before first use, and explains how to unlock the model instead of showing a raw download failure — thanks paoloantinori! (#1442)
The repository moved to github.com/debpalash/VoiceStudio. Every link in the app, docs and scripts now points there; GitHub redirects the old URLs, and the Docker image paths, the app bundle identifier and your data folder are all deliberately unchanged. (#1394)
The app is now VoiceStudio (previously OmniVoice-Studio). Only the name you see changes — your data folder, settings and the Docker image paths stay put, so upgrading needs nothing from you. On Linux the .deb is now voicestudio; remove the old omnivoice-studio package once.
macOS floor raised to 13.3 (Ventura) — the frontend has required Safari 16.4 for some time, so macOS 12 was a promise the stack could not keep (#1268)
The first-run setup screen no longer overpromises. It claimed "no account, no cloud, no telemetry" without qualification — untrue for anyone who opts into analytics — and now says what actually holds either way: your voices, recordings and projects never leave the machine, and no processing happens in the cloud.
Dictation no longer shows a floating pill. The hotkey records, transcribes and pastes with nothing on screen; the tray icon still marks recording, and anything needing your attention (Accessibility, microphone, a failed transcription) now arrives as a notification in the main window.

Added

Gallery personas preview through the local backend, keep their full voice-design recipe, and open directly in Voice, Stories, or Audiobook — and can be saved as local profiles with validated audio references. (#1542)
The demo audio the app has always advertised now actually ships: previews for all seven voice-design presets, the three dictation replay clips, and the dubbing demo's source video plus four dubbed languages with subtitles. Every one of those was a dead link before — the tooling that renders them required macOS, so on Windows and Linux the files were never built. (#1517)
Demo assets are rendered by VoiceStudio's own engine, so the tooling runs wherever the app does, and the demos are made by the thing they demonstrate. (#1517)
A machine can now join a control plane from the app: Settings → System → Remote workers → Lend this machine's GPU, paste the join code, done — no environment variables and no restart. The address travels with the code, so the machine reconnects on its own afterwards. (#1516)
Join codes and connection strings are shown as a QR code alongside the text, with a live expiry countdown — scan it from the other machine instead of retyping forty characters. (#1516)
A Compute control in the status bar: pick local or a remote machine, turn remote workers on or off, and mint a join code without opening Settings. It appears only once you have opted in or enrolled a machine. (#1516)
A worker waiting for approval can be approved from its row. The panel labelled that state before but offered no way out of it. (#1516)
Model Catalogue — a workspace of its own for engines and models: browse every TTS, transcription and LLM engine with its device routing and install state, pick the default for each, and install or remove model weights, all from one screen instead of two Settings categories.
Remote GPU machines can now accept connections instead of dialling out, so several people can use the same box at once — each gets their own revocable connection string, with certificate-pinned TLS, a live list of who is connected, and a disconnect button. (#1496)
Remote GPU model downloads now use the normal Models install flow and show per-worker progress. (#1478)
Settings → System → Remote workers sends individual jobs to GPUs on your other machines while everything else stays here. Off by default; each machine is added with a single-use token and approved before any audio reaches it. See docs/remote-workers.md.
First-run setup now recommends a screen-aware interface scale, with compact controls available throughout setup. (#1502)
OrcaRouter is now available as a named OpenAI-compatible LLM provider — thanks Marc-oss-hub! (#1499)
IndexTTS 2.5 is available as a pinned one-click sidecar with five-language dubbing, expressive cloning, and backward-compatible IndexTTS-2 support. (#1482) — thanks marwanlhabti5-coder!
Voice recording now offers microphone and channel selection with a live input-level meter on every desktop platform. (#1481)
Settings → Appearance → Navigation style switches the workspace switcher between the icon rail down the window edge and browser-style tabs across the title bar. Both offer the same workspaces; the choice sticks across launches, and the rail stays the default. Tab labels fold down to icons when the title bar runs out of room — the workspace you're in keeps its name. (#1412)
Portable mode lets you choose the folder — press Change… on the first-run setup screen and put the whole install on an external drive. It also stops being greyed out after a default Program Files install. (#766)
Settings → Privacy now has an Invisible watermark toggle. On by default, available to everyone, and it only affects audio generated after the change. (#1308)
A new opt-in crash-isolated TTS engine, so a native crash takes down the sidecar instead of the whole backend — thanks paoloantinori! (#1292, #1298, #1304)
PocketTTS (Kyutai), an opt-in CPU-only engine for fast, low-latency renders in six languages (en/fr/de/pt/it/es) with zero-shot cloning from a reference clip. Enable in Settings → Engines — thanks paoloantinori! (#1306, #1328)
A warning before a slow generation, rather than after a five-minute wait. (#1280)

Docs

Engine acceptance: new docs/engine-acceptance.md documents the job map, the bar a new engine must clear, and the out-of-tree path (#1306)
macOS install notes and the README support table now state the real floor (#1268)
Contact: the project X account is listed alongside Discord (#1313)
OMNIVOICE_ALLOWED_ORIGINS is finally documented: a browser loading the UI from another machine's origin needs the backend's CORS allow-list, which neither server mode nor trusted networks touches — thanks vanderlpp! (#1348)

Fixed

AMD/ROCm hosts no longer crash ASR with "CUDA driver version is insufficient": ROCm torch reports itself as CUDA, but whisperx/faster-whisper run on CTranslate2, which is NVIDIA-only — they now take the CPU path there, and auto-detect prefers pytorch-whisper, which genuinely uses the HIP GPU. (#1529)
Crash reports now carry the crashed run's own stderr: the shared error log is append-only with per-run offsets, so a restart can no longer overwrite the dying process's final output with the replacement's healthy startup. (#1510)
Wayland: a stale portal identity no longer kills the dictation shortcut for the whole session. The desktop entry the app writes for the GlobalShortcuts portal could point at a binary that has since moved (a cargo clean, a relocated AppImage) — GNOME then refuses the bind with "App info not found" and the hotkey silently dies. The entry is validated and rewritten at startup now. (#1526)
The guard that keeps transcription on the degrading ASR loader now scans the whole backend, not just the routers — a service that transcribes on a request's behalf skipped ensure_loaded() just as thoroughly. (#1519) — thanks ahov520!
The Linux app icon is no longer blank. Every AppImage since v0.4.2 shipped .DirIcon as an absolute symlink into the machine that built it (/home/runner/work/…), so the link dangled on every user's computer and file managers, app menus and desktop integration all drew nothing. The release build now verifies the icon resolves inside the bundle before publishing. (#1518)
The Linux desktop entry no longer ships an empty Categories=, which desktop-file-validate rejects and menu builders skip. (#1518)
Wayland: the dictation shortcut now actually starts dictation. The desktop portal registered the key correctly — GNOME and KDE even showed it back — but every press was discarded while decoding the compositor's signal, so the hotkey did nothing on any Wayland session. (#1490)
The first-run "Choose a comfortable UI size" screen no longer stutters while you sit there. Applying a scale resizes the window's own viewport, which the screen was reading back to re-pick a size — so it flipped between two sizes forever without anyone touching it. (#1514)
Transcription now moves to the next working engine when the auto-picked one passes its availability check but breaks on first real use, instead of returning an internal error — the recovery dubbing already had. Affected accurate-mode transcription, the OpenAI-compatible API, batch, dub verify, and voice-clone reference text. (#1512)
A malformed request now gets a clear 422 instead of an internal error, and uploading a file to an endpoint that expects JSON no longer copies the whole upload into the app log — a 145 KB clip wrote roughly 500 KB of log, recording your audio in the file people paste into bug reports. (#1513)
The Simplified Chinese (zh-CN) translation no longer mistranslates brand names and technical terms — Discord, Tailscale, Hugging Face, IPA, and LLM (Cinematic) were rendered as nonsensical literal translations, and ~250 more awkward machine-translation strings are now natural Chinese. (#1508) — thanks anyingiit!
Worker restart coverage now waits for the registration response to persist its identity instead of racing the client callback in CI. (#1505)
Dub language and export selections now restore without false schema warnings, and remote-worker port 7443 is identified instead of reported as a generic timeout. (#1504)
A configured remote backend now bypasses local first-run setup, verifies itself before app requests begin, and shows recovery instead of leaving the desktop stuck on Setup. (#1503)
An idle voice model now actually hands its memory back. The unload emptied the GPU cache a moment before releasing the model, so it freed nothing while reporting success — a GPU machine lending its card sat on 3.6 GB indefinitely. (#1495)
Unloading a model on an NVIDIA GPU now returns the last ~770 MB too. A single 8.5 MB cuBLAS workspace sat inside the model's memory block and kept the whole block reserved, so an idle machine held 1.2 GB instead of 470 MB no matter how often you pressed Flush Memory. (#1495)
Flush Memory reports reserved GPU memory alongside allocated. Allocated alone reads near zero right after an unload while the GPU still shows gigabytes, which is exactly the case people were reporting. (#1495)
The AudioSeal watermark models are released after the same idle period as everything else, instead of staying in memory for the life of the app once anything was watermarked. (#1495)
Remote GPU workers now synthesize a dub's fresh segments as one coarse job with live progress and cancellation; fitting, assembly and RVC remain local. (#1478)
Gallery voice previews now fall back to a local render when a downloaded clip cannot be decoded, instead of failing silently. (#1478)
A second VoiceStudio instance can no longer silently share the remote-worker port; it keeps running locally and explains how to resolve the conflict. (#1478)
Remote GPU jobs stay pinned to the selected worker across retries and restarts, stop when their caller leaves, and cannot return from cancellation as completed. (#1478)
Remote GPU model labels now survive registration, legacy blank model IDs share one capacity slot, long jobs retain bounded leases, and idle cleanup cannot evict a live local render. (#1478)
Remote GPU jobs now stop before dispatch when that worker lacks the model, offer the download there, and refresh scheduling as soon as it finishes. (#1478)
Leaving a screen while its waveform is still loading no longer opens a bug-report prompt for a normal cancelled request. (#1498)
An unreachable remote backend now opens a retryable recovery screen instead of sending the app into local model setup, with clear TLS, CORS, network, HTTP, and wrong-port guidance — thanks debpalash! (#1501)
Linux production test launches now stop their own extracted AppImage before resetting SQLite and logs. (#1494)
Restored the pre-release version to 0.4.2 while the next release remains in preparation. (#1488)
Large multi-language dubbing batches now use compact searchable language and track managers instead of overflowing the editor. (#1492)
Dictation shortcuts now register and rebind through the desktop portal on Wayland, honor custom keys in focused app views, and show the effective platform keys. (#1490)
Multi-language dubbing now translates, edits, generates, retains, and exports every selected language, and its language picker stays visible at viewport edges. (#1486)
Dubbing's From video cast now uses available source-audio samples for every speaker and short line, including jobs without a pooled diarization clone. (#1484)
Basic Dubbing translation remains available without an LLM; Cinematic and Autofit now degrade through the existing Fast translation path instead of blocking the quality choice. (#1481)
Linux microphone recording now falls back to WAV when WebKit cannot encode MediaRecorder audio, and desktop scaling/titlebar controls remain responsive at every UI scale. (#1481)
Dubbing can install a missing ASR model and retry the same job, navigate back through completed stages, and finish transcription under low GPU memory without producing an empty transcript. (#1481)
Filenames and other outside data can no longer forge extra lines or terminal commands in backend and frontend diagnostic logs. (#1457)
Backend journal, dictation reset, voice-catalog, and crash-notification failures are now visible and retryable instead of being silently ignored. (#1459)
Backend failures keep raw tracebacks, local paths and credentials in the local log instead of returning them in API responses. (#1454)
GPT-SoVITS connections now stay on loopback or explicitly trusted networks and cannot escape through redirects or DNS rebinding. (#1463)
Engine discovery no longer exposes probe exceptions, local paths or credentials in API responses and logs. (#1460)
Failed gallery, batch-video, and desktop-log cleanup is now reported instead of silently claiming success, and diagnostic redaction fails closed if a scrubber breaks. (#1458)
Remote backends can no longer probe or overwrite arbitrary host files through native-only tools, and imported or persisted paths cannot escape their VoiceStudio data folders. (#1455)
Linux releases now verify that the AppImage actually contains the compatibility launcher, instead of silently shipping Tauri's stock launcher and opening as a blank window on newer Mesa systems. (#1464)
Patched dependency releases now cover 35 Python and Rust security advisories without weakening VoiceStudio's GPU or offline-runtime compatibility. (#1456, #1472, #1473, #1474, #1475, #1476, #1477)
Curated models now install and repair from reviewed, immutable revisions; custom MOSS remote code requires an explicit safety opt-in. (#1453)
YouTube imports that require a signed-in session can now use an explicitly selected cookies.txt export for one import; VoiceStudio never reads browser cookies silently and makes two best-effort attempts to delete its temporary copy. (#1429, #1432) — thanks dongqing1968-sudo and @phamvandu9595-tech!
First-run source builds no longer stop after uv was successfully downloaded just because its installer failed during a later shell-profile step; app-private uv installs no longer touch shell profiles at all. (#1438) — thanks AdrianoCahete!
Model files damaged by an interrupted download now repair themselves instead of failing every generation, including invalid config.json files and corrupt weight headers. — thanks overrunau and @zherunh! (#1406, #1437)
ROCm Docker now installs and starts the backend with the same Python whose AMD torch build was validated, instead of launching a second CUDA-only environment and silently running on CPU. (#1274) — thanks simmessa and @spicchio72!
An error whose text merely contained the digits 401 — a file path, a byte count, a job id — no longer tells you to fix your Hugging Face token. (#1427)
Custom MLX model IDs and saved voice instructions are now validated in bounded time, so malformed input cannot stall the backend. (#1446)
Streaming and provider failures now return stable recovery guidance without exposing exception details. (#1462)
Server-mode settings mutations require the admin API key, while host destinations and executable paths can only be selected through the native desktop app. (#1448)
Automatic model-mirror checks now reject untrusted URLs before opening a network connection. (#1447)
Sidecar engines no longer break when a library they load prints to the console. Those bytes landed in the middle of the engine's data stream, failing the generation and leaving the connection scrambled for every request after it. (#1428) — thanks 1335-Group!
A generation abandoned while stuck on an internal lock now says so, instead of blaming your hardware and suggesting shorter text. Nothing had been computed, so none of that advice applied. (#1416, #1419)
A machine with a GPU that ends up on CPU now says why — a missing device node, a permissions problem, a card newer than the installed ROCm, an HSA_OVERRIDE_GFX_VERSION that is doing more harm than good, or an NVIDIA driver the container can't reach each read differently. Before, all of them looked identical to having no GPU at all. (#1274, #1228)
The first generation on an engine that still has to install itself no longer gives up part-way. The install reports progress now, so the generation waits for it instead of hitting its own five-minute limit. (#1414)
A slow machine is no longer told its IndexTTS-2 install isn't there. The check that confirms an engine's virtualenv gave up after 10 seconds and counted that as a broken install, so a cold first run 500'd; it now waits longer and treats slow as unproven, not broken. (#1414) — thanks OracleNightmare!
A broken Python environment now says so, instead of blaming the app's own install. A missing or mismatched torch/transformers surfaced as "omnivoice not importable" and sent people reinstalling the wrong thing. (#1415)
A model that fails to load at startup no longer leaves the app looking healthy while producing nothing — the failure and its remedy now show up in the model status. (#1415)
Generating with the default engine works again on everything built from main since the rename — source checkouts, preview builds and Docker :latest all run the same backend, whose model import had been rewritten to a class name the library doesn't export, failing every generation with "cannot import name 'VoiceStudio'". The class keeps its library name, and a guard test now pins it. (#1420)
Running from source no longer dies at startup when a database migration is pending. Alembic resolved the migrations folder relative to wherever the app was launched from — fine from the repo root, fatal from the desktop shell (tauri dev), which reported "Path doesn't exist: backend/migrations" and stopped. The path is now anchored to the repo, wherever you start it. (#1420)
The first generation after startup no longer stalls or 500s while the model is still loading. A cold load reached from a worker thread waited on a lock owned by a different event loop, which either errored outright or deadlocked until the job was abandoned. (#1417)
The voice-design model on Apple Silicon works again. Its description was being dropped before it reached the engine, so every generation failed with a raw 400 no matter what you typed. (#1405)
The first-run setup screen no longer times out while it waits for you. Taking more than two minutes to choose an install location, region or mirror made the app declare "Setup failed", and Retry landed back on the same screen with the same clock — so a first install could never be completed. (#1376)
Transcription on an NVIDIA machine whose cuDNN 8 libraries are missing no longer kills the backend outright. The app checks the library before picking a transcription engine and falls back to PyTorch Whisper, instead of handing off to a component that aborts the process with no error and restarts into the same crash. (#1371)
The dictation model picker now tells the truth about download size. Every one of the seven models was wrong: Parakeet TDT v3, the recommended default, said 180 MB and actually downloads 670 MB, while the small low-RAM fallbacks were advertised as three times bigger than they are. (#1398)
Dictation with the 0.6B Parakeet models is steadier under load — they now decode on more threads (still capped by your CPU, still overridable with OMNIVOICE_SHERPA_ASR_THREADS). The small models are unchanged. (#1398)
The dictation hotkey no longer leaves a blank dark square stuck on your desktop. A press that arrived while the pill was re-arming was dropped, and the window it had already opened had nothing in it and no way to close it. (#1398)
The blank dark square is gone for good: the dictation window could mistake itself for the main window when its shell wasn't ready yet, and once it did, nothing in the app could close it again. It now learns which window it is before any of its code runs. (#1398)
Dictation is more reliable to trigger: the hotkey listener no longer briefly detaches every time the pill changes state, so a press is never silently lost. (#1398)
An auto-captured crash report now keeps the error that actually caused the crash. Python prints a chained traceback oldest-first, so trimming the log to its newest end kept the generic wrapper and cut the real cause — the reports that needed the detail most were the ones that arrived without it. (#1376)
Text ending in punctuation no longer wastes a whole synthesis pass on it. A chunk boundary could leave a trailing fragment with nothing speakable in it, which the engine renders as nothing at all. (#1330)
A take that is missing part of your text now says so instead of coming back quietly short. When the engine renders a sentence to nothing, the app names the missing text and suggests re-generating — until now the only way to notice was to read along. (#1330)
A long render on modest hardware is no longer abandoned as "too heavy for the available compute" while it is visibly working. A generate that keeps finishing chunks now extends its own deadline (bounded), the way a model download already could; one that stops producing anything still fails on time. (#1338, #1348, #1391)
A backend that dies while loading its own Python dependencies is no longer reported as a memory problem. The crash notice now says the environment is incomplete and points at "Clean & Retry", instead of sending users to flush a model that had nothing to do with it. (#1282, #1376)
A UI whose API requests land on the wrong host — a rehosted frontend, or a reverse proxy with no API route — no longer echoes that host's raw 404 page as the error. It now says the responding server is not a VoiceStudio backend and points at the Backend URL setting and the proxy route. (#1385)
Building the GGUF engine from source produced a binary that died on its very first spawn ("libggml.so.0: cannot open shared object file") — the build script deleted the shared libraries it had just linked against. It now ships them next to the binary on every platform, and the backend puts that folder on the loader path — thanks vanderlpp! (#1348)
The GGUF engine's hard 120-second per-render kill switch — which was reaping legitimate CPU-only renders mid-synthesis — is now 600s, tunable via OMNIVOICE_GGUF_GENERATE_TIMEOUT_S, and the timeout error names that setting — thanks vanderlpp! (#1348)
Every subprocess TTS engine would have turned a stereo render into noise: the mono downmix always averaged axis 0, which is time rather than channels for channels-last audio. Unreachable today since every engine returns mono, fixed in all five before it isn't. (#1328)
First-run wizard: the Continue button and the Hugging Face token box were pushed below the window with no way to scroll to them — a layout container grew to the full model list's height, defeating every scroll clamp inside it. The pinned row now stays on screen at every UI scale, with the model list scrolling under it. (#1382, #1383)
Dubbing the same video twice no longer ties the second job's cloned voices to the first job's files — deleting the older dub from history was silently turning the newer one's single-segment regens into a default voice. (#1331)
...and deleting a dub whose files an existing saved dub still renders from now keeps those files on disk (the history entry still disappears) — protecting dubs created before this fix, whose references already cross directories. (#1331)
An unclean previous shutdown is no longer announced as a crash: the notice says what it actually knows, names the benign causes (sleep, force-quit, a stopped VM), and the one-click bug report is only offered when there is evidence to put in it — an empty report helps nobody. (#1375)
A first-use generate no longer fails at 300s while its model is still downloading: the download's own progress heartbeats now extend the generation budget (bounded), so a slow connection isn't reported as too-slow hardware. A job that goes silent still dies at the original deadline. (#1367)
A generation that hits its time limit now says so, instead of "an error VoiceStudio doesn't recognize" followed by an empty TimeoutError:. It names the likely causes and the setting that raises the limit. (#1368)
The "transformers install is incomplete" advice now names torchvision — the package whose version mismatch actually produces that error — and points at the pinned reinstall that repairs it, instead of a reinstall that left the broken package untouched. (#1376, #1357)
A model download cut off mid-request is no longer reported as a broken transformers install — reinstalling could never have fixed a dropped connection. (#1347)
A TLS connection cut during generation is explained as the dropped download it is, instead of falling through as an unrecognized error carrying _ssl.c:1016. (#1335)
Windows "paging file is too small" no longer suggests the Flush button, which cannot help. It now names the virtual-memory setting to change, and says plainly that it is not a network problem. (#1334)
A port conflict that resolves itself while the backend is dying no longer reports a bare "Backend died (exit code 1)" — the conflict is named even when the other process has already let the port go. (#1364, #1223)
Fresh installs failing to import transformers.HiggsAudioV2TokenizerModel with "RuntimeError: operator torchvision::nms does not exist" are fixed by pinning torchvision==0.23.0 to match torch 2.8.0 — thanks HanzlahCh! (#1358, #1357)
...and that pin now actually reaches Colab and Docker: both install with uv pip install, which ignores the pyproject setting the pin lived in, so the torch trio could still drift apart. It is passed explicitly now. (#1357)
Every RTX 40-series card (4060–4090) was declared unsupported and silently run on the CPU. The compatibility gate demanded an exact sm_89 match, but PyTorch ships sm_86 kernels that already cover Ada. (#1285)
Under-provisioned hardware is now flagged before a synthesis starts instead of after the full compute budget expires. (#1240, #1246, #1248, #1277, #1283, #1284)
Long text on a CPU-only machine gets the same warning up front. (#1260, #1299)
A crash inside the compute stack no longer blames VRAM: a segfault or Windows access violation now points at the GPU driver or an incomplete model download. (#1275, #1293)
ffmpeg failures report the failure instead of ffmpeg's build configuration. (#1309)
A cut TLS connection is explained in words rather than as _ssl.c:1016. (#1301)
torch.compile is skipped when the torch library path contains a space, instead of failing in the linker on every load. (#1266)
macOS Preview updates work again — the updater bundle had been colliding with itself since early July. (#1281)
macOS Preview updates no longer fail signature verification — the preview manifest is rebuilt from the published assets and every signature in it is verified against the file it points at — thanks Pinkers01! (#1327)
A dub whose transcription stream is cut by a reverse proxy now says so, instead of blaming the ASR model. (#1317)
The dev backend going quiet under --reload is named as auto-reload rather than reported as a crash. (#1261)
Building from source: bun run desktop-prod:run, documented as a re-launch, wiped the app's data every time — voice profiles, projects and outputs included. It now keeps them — thanks Kakuzen93! (#1333)
Audiobook: a chapter that fails to render now shows the reason in the chapter list and in the final error, instead of a red row whose cause existed only in the backend log — thanks Reaksa-Cambodia! (#1321)
Audiobook: an engine that stops without producing audio no longer stalls the render forever with no error and no timeout. (#1321)
Linux AppImage: a permanently blank window on Mesa 26.1+ hosts (Arch/CachyOS and other rolling distros) — the bundled WebKit ran against a newer system Mesa than it was built for, and no environment variable could help because the failure precedes every rendering flag; the launcher now lets a newer system WebKitGTK take precedence — thanks rvasilev and @HannaLovvold! (#1258, #1244)
Linux AppImage: OMNIVOICE_PREFER_SYSTEM_WEBKIT=1 forces your own WebKitGTK for hosts where its version can't be read automatically (no pkg-config), and =0 forces the bundled one (#1258)
Dubbing: the transcription overlay said "Transcribing with Whisper…" whatever ASR engine was actually running — it now names the stage, in all 21 languages — thanks paoloantinori! (#1352)
Error messages no longer arrive with terminal colour codes spliced into the sentence (download: ^[[0;31mERROR:^[[0m …) — every surfaced failure is cleaned now, whichever tool produced it. (#1344)
Linux AppImage: recording failed with "No microphone found" on hosts whose GStreamer is newer than the build runner's, even with a verified-healthy audio stack — your own GStreamer now takes precedence, and the plugin cache is app-private so it can neither be confused by nor corrupt the one other apps use — thanks Kakuzen93! (#1333)
Linux AppImage: that GStreamer preference actually takes effect — the check guarding it could never pass, so it had been silently doing nothing. (#1333)
A TTS job abandoned for exceeding its compute budget now records where it was actually stuck, so a hang stops being reported as a machine that is merely too slow. (#1338, #1329, #1348)
Translation through LM Studio works. The built-in model name was the placeholder local-model, which LM Studio rejects because it serves whatever you have loaded — VoiceStudio now asks it, and a 404 from a local server names the models that ARE loaded instead of telling you to check a URL that was fine — thanks biga73! (#1332)
Generation that silently dropped the end of the input now says so. When an engine returns no audio for part of the text the result sounds clean and is simply short, so the only way to notice was to read along; the backend log now names the sentences that produced nothing. (#1330)
Dubbing: a re-rendered line that quietly came back in a default voice instead of the cloned one now says why in the backend log — the clone clips are extracted per job and a saved dub outlives them, so regenerating after cleanup loses the reference with no error. (#1331)
RTX 40-series GPUs are used again instead of being sent to the CPU. (#1289)
Apple Silicon: transcription no longer needs a system ffmpeg, as the docs always said — thanks gambletan! (#1436)
A failed audiobook chapter says why, instead of turning red and saying nothing. (#1325)

CI

Windows CI falls back to a static ffmpeg build when the Chocolatey feed is down, instead of failing the run. (#1542)
The stdio wire protocol every engine sidecar speaks is now tested once across all nine of them, instead of against a single engine — a bug in any one sidecar's copy gets caught — thanks paoloantinori! (#1408)
Windows smoke tests stopped silently passing a broken ffmpeg install, and every smoke leg is now budgeted for a cold dependency install. (#1290)
Test suites no longer leak config paths or model-manager shutdown state into one another, which had been failing unrelated pull requests. (#1269)
The nightly preview build stopped refusing to publish its own healthy updater manifest when the macOS legs finished a few minutes ahead of the slowest one — Preview-channel users were silently left without new builds.

macOS Intel artifacts

2c640c2db2c4144ece142acfd1d83f636a2d3495fbd98c4e385605794936c1a8  VoiceStudio_0.5.0_x64.dmg
207c6f0077978416df266966431f434a735e6cfa9ac229015db7d4a85ec67d2e  VoiceStudio.app.tar.gz
30e20278d5f5712858b598a4d50fbb9130a3cfb6b356bc5153b3c97747e1cbd6  VoiceStudio.app.tar.gz.sig

Windows x64 artifacts

5b4d6e136f90bee4544e7b5d595c31bf6a0e2eab30d79ad950f799dd29ecf864 *VoiceStudio_0.5.0_x64_en-US.msi
e9561376c820580290ad19f89526d1187c4383c08c7f2a7d629dbd1d464d6392 *VoiceStudio_0.5.0_x64_en-US.msi.sig

Linux x64 artifacts

1a26694cebd461592c8ab7b9c1fb2ec2c820e47461c58d824dd6fe3c487e2cee  VoiceStudio_0.5.0_amd64.AppImage
7b1ea9dbe0efa2f2bdfb1d97a329f761e0bad03784b175741980cf7b8ef3b267  VoiceStudio_0.5.0_amd64.AppImage.sig

Contributors

Thank you all 💜

详细ChangeLog

v0.4.2

2026年07月28日

Highlights

The update prompt is a small toast with buttons, not a screenful of release notes
Installing an update no longer throws away work that is still running
Quitting the app mid-generate stops reporting itself as a crash
A half-downloaded model repairs itself instead of dead-ending
"Dismiss" no longer reads as "terminate an employee" in five languages

Changed

An available update now announces itself as a toast with Install and restart, What's new and Later, instead of only a dot beside the version number. The release notes stay in Settings → Updates, where there is room for them — a version's notes are the whole changelog section, and rendering them inline is what made the old prompt fill the screen (#1272)

Fixed

Installing an update no longer relaunches the app while work is running. The check only knew about dub synthesis, so a restart could silently discard an upload, a transcription, a translation, an export or a standalone synth — and two overlapping synths used to cancel each other's protection. Install is now greyed out while anything is in flight (#1272)
A half-downloaded model now repairs itself instead of failing with a raw 500. The automatic repair recognised only one of the two ways the loader reports missing weights, so an interrupted download whose subfolder failed to load got neither the repair nor a hint about what to do (#1273)
Quitting the app with a generate queued reported "500 Internal Server Error: model load skipped: backend shutting down" and offered to file a bug for it. A shutdown is not a fault: the backend now answers 503 with what to do, and no bug report is offered for it (#1276)
Dub history: clearing a large history while a render was running could still resurrect the deleted job — which markers survived depended on the process hash seed, and an oversized purge could discard a live one (#1252)
German, Japanese, Russian and both Chinese locales rendered "Dismiss" as the employment sense — "terminate an employee" — on close buttons (#1272)
The "wait for the current job to finish" message named dubbing specifically, though it now covers uploads, transcription, translation, exports and synthesis; reworded across all 21 languages (#1272)

Linux x64 artifacts

81c1265984ec4376aecd28417413b037cdef3b1b9fe6bc7cb8d731cad7811f60  OmniVoice Studio_0.4.2_amd64.AppImage
4e880f64a85333b9c54a1a15990e54b02355667faf609e0b82160e75d88d8906  OmniVoice Studio_0.4.2_amd64.AppImage.sig

macOS Intel artifacts

22f0212893ee2b0b373329b730520fad943bd5685d1a6219a4b4b53b1e8ef949  OmniVoice Studio_0.4.2_x64.dmg
7e1ebf185e456fe89a5fbee3603f1992604cf9c672717fe7d63a0fa64d2d14e1  OmniVoice Studio.app.tar.gz
e2d5cc678b4ed3ee637e37a48d6e109babe6d0311d041dc867e3bc85493b160a  OmniVoice Studio.app.tar.gz.sig

Windows x64 artifacts

6cd5b924fa1e6ec2022ff3db0c18bec24e765fb71a294f9101fac807d4a132a1 *OmniVoice Studio_0.4.2_x64_en-US.msi
1ebddf5fcd019345085763f127a6be20b6d3cff0b241a2095e1aafd987b4d369 *OmniVoice Studio_0.4.2_x64_en-US.msi.sig

Contributors

Thank you all 💜

详细ChangeLog

v0.4.1

2026年07月27日

Highlights

AMD GPUs are used again — every ROCm host was silently running on the CPU
Two synth failures that used to say "an error OmniVoice doesn't recognize" now say what actually went wrong
A dub URL ingest that fails on a disk problem now says which folder and why
A broken audio dependency no longer takes the whole backend down at startup
A GPU too small for the chosen engine now says so up front, not after a five-minute wait
A port conflict now says so, instead of "Backend died (exit code 1)"
A model download that dies at 90% now resumes instead of failing the install
First run: Continue and the Hugging Face token box no longer sit under the status bar
macOS 12 (Monterey): the app launches again instead of dying on startup
Exporting a voice or a dub no longer fails when the name isn't spelled in Latin letters
Two more failures that used to arrive as raw OS text now say what to do about them
Unload works on every model the panel offers it for, and a language the active engine can't speak says so
Deleting a dub no longer un-deletes itself when the job it belonged to finishes

Changed

First run: the status bar (Logs, version, Sponsors) appears once you reach the studio, instead of overlaying the setup steps (#1241)

Added

OMNIVOICE_MCP_ALLOWED_HOSTS — comma-separated host patterns (e.g. host.containers.internal:*,192.168.1.5:*) that extend the MCP SDK's DNS-rebinding allowlist, so AI agents running in Docker containers or on other machines can reach the /mcp endpoint. The SDK default is localhost-only; this env var is opt-in (#1249)

Docs

Docker: ROCm section explains that torch.cuda.is_available() == True isn't proof the app is on the GPU, and notes the --group-add needed for /dev/kfd on rootless hosts (#1228)

Fixed

Deleting a dub while it was still importing crashed the import with the toast ingest: 'mgw39lx3' — a dict key and nothing else — and the delete could then be undone by the job's own pending write, in history or mid-render; both are fixed, and no failure can present itself as a bare value again — thanks dustmaker124-ui! (#1252, #1253)
macOS 12 (Monterey): the app threw on startup and never started the backend — it called a Safari 16 method on the WebView that macOS ships. It launches and works now; some styling still needs a newer WebView (tracked in #1268) — thanks singhrahat! (#1245)
Settings → Engines: Unload failed with 400 Unknown model id: engine:kittentts on any in-process engine — the panel offered the button for ids the backend never accepted; the warm dictation model had the same gap — thanks JavaxmI! (#1247)
Picking a language the active engine can't speak recited 23 codes without saying which engine refused or that switching engine was the fix — thanks pulananave! (#1257)
A YouTube import that failed as "DRM protected" and then worked on a manual retry now escalates the player client automatically, and a genuinely undownloadable video says so — thanks gysahlgreene! (#1254)
Exporting a voice profile, persona, dub, subtitle or stem whose name is Chinese, Japanese, Korean, Cyrillic, Greek, Hebrew or emoji failed with a 'latin-1' codec 500 — every download endpoint now sends the name correctly, and browsers get the real one back — thanks zvxzdx! (#1262)
A synth that failed because ffmpeg/ffprobe wasn't on the system path said "an error OmniVoice doesn't recognize"; it now names the media engine and points at Settings → Audio tools, and the app's own copy is published on PATH so dependencies find it in the first place — thanks Heuvelsma! (#1256)
Windows "The paging file is too small" arrived as a bare 500; it now explains that this is a virtual-memory setting, not full RAM, and gives the steps to raise it — thanks trankeny545-sudo! (#1251)
AMD/ROCm: every ROCm host was silently force-routed to the CPU — the compatibility gate compared a CUDA sm_ tag against a ROCm build's gfx list, which can never match — thanks simmessa! (#1228)
AMD/ROCm: torch.compile was disabled on all AMD hosts by the same mismatched comparison (#1228)
AMD/ROCm: HSA_OVERRIDE_GFX_VERSION is auto-set only when your card genuinely needs it and the remap target exists in your build; gfx1150/gfx1151 (Strix Point/Halo) added to the map (#1228)
Windows blocking an engine file (Smart App Control, WDAC, or AppLocker) is now named, with the fix for personal and managed PCs — thanks AdityaHemantBhat! (#1227)
A failed audio write (LibsndfileError: System error.) now names the target file, its folder's writability and the drive's free space — thanks morozov28061995-boop! (#1221)
Dub URL ingest: a disk error now names the job folder, its writability and the drive's free space, instead of pointing at the system TEMP folder it never used — thanks dustmaker124-ui! (#1225)
Dub URL ingest fails immediately when the job folder is missing or unwritable, instead of starting a download that can only fail (#1225)
The backend no longer dies at startup when transformers can't resolve its audio tokenizer (a missing or mismatched torchaudio, common on Google Colab) — it starts, and the error arrives with a repair hint — thanks Navdeep-Chauhan-777! (#1229)
Importing omnivoice.utils.* no longer drags in torch, torchaudio, transformers and the full model definition — thanks Navdeep-Chauhan-777! (#1229)
Colab notebook: the install cell now catches a broken environment with the real error, instead of a 5-minute health timeout two cells later — thanks Navdeep-Chauhan-777! (#1229)
A GPU with less VRAM than the chosen engine needs is flagged in Settings → Engines before you generate, instead of showing a clean green "accelerated" until the job times out — thanks AdityaHemantBhat and @beingavais! (#1226, #1222)
A generation timeout now names your actual card and its VRAM and recommends a lighter engine (#1226, #1222)
First run: Continue and the Hugging Face token box rendered underneath the status bar, off the bottom of the window — the wizard laid itself out against the viewport instead of its own frame (#1241)
A busy port 3900 now reports a port conflict instead of "Backend died (exit code 1)", in every language — thanks xipb14! (#1223)
The app verifies it actually freed the port before starting the backend, rather than assuming the kill worked (#1223)
A model download truncated near the end is now retried and resumed instead of aborting the whole install — thanks Reaksa-Cambodia! (#1224)
Engine first-use downloads (VoxCPM2, MOSS-TTS-Nano) retry transient network failures instead of failing the load outright (#1224)
A backend killed by the OS mid-stream now leaves a low-memory trail in the crash report (#1224)

macOS Apple Silicon artifacts

3c88dc39fbef840a81d6c4d5df862d4cb1970bebe320da973e7f304d97e4febe  OmniVoice Studio_0.4.1_aarch64.dmg
be702e7f03913da7a4ff4bb0049dc2dd225d3c5602d959a6dcbcce7330c75049  OmniVoice Studio.app.tar.gz
3845c8a23fc79b8025cd8e3a6b8766901c0d1928ef6e0569e0ee6c25b63a541c  OmniVoice Studio.app.tar.gz.sig

Windows x64 artifacts

d6d407fb26ce2aa7c474909d6cbce236daa044a56e258c46681296ef9c968d00 *OmniVoice Studio_0.4.1_x64_en-US.msi
011c4c8e9c58bf2de3b06a3a76df26718b5c6f0ecb168bd3d495c3b334952e96 *OmniVoice Studio_0.4.1_x64_en-US.msi.sig

macOS Intel artifacts

91fd008c1bbd60b2f665ef7836a326c1756fee5b0cf3db15b9c845d249431f0b  OmniVoice Studio_0.4.1_x64.dmg
84a8b1ac1da95ecdfb1c0be08cb5a4e57ca5cead3e0a14a416caf50e15aec6d7  OmniVoice Studio.app.tar.gz
82e6bb094852c15ebfae54f45c22ac4185d36bd398cf52ac3e1e1c73cf0b26ca  OmniVoice Studio.app.tar.gz.sig

Contributors

Thank you all 💜

详细ChangeLog

v0.4.0

2026年07月21日

Highlights

Audiobooks, end to end — a real Stop with live per-chapter progress, a multi-voice cast, expressive controls, a markup toolbar, live stats, and a one-click sample
Pick a designed voice from the Gallery anywhere you choose a voice — audiobook, Stories, and Dubbing
Dub Paste Translation — drop in a translation or .srt and it maps straight onto your segments, timings intact
Downloading a finished audiobook no longer hijacks the app — it just saves
First run is ~2.4 GB, not ~5 GB — only the TTS model is required; ASR picks are curated per platform
Guided mic + Accessibility permissions with Open Settings deep-links; Parakeet TDT v3 on Apple Silicon
Opens in your system language, with a one-tap switch back to English
Security: server-mode admin routes can't be reached by a trusted-network client without the API key
A render error shows a recoverable card instead of a blank window; queued and long generations stop failing with a bogus "too heavy for your hardware"

Changed

Settings → Models: grouped catalog (TTS / ASR / Dictation / Diarisation), "recommended for this machine" chips, incompatible models collapsed behind a toggle
Only the TTS model (~2.4 GB) is required on first run; ASR picks are curated per platform via curated_on in models.yaml (MLX on Apple Silicon, CT2+Turbo on CUDA, PyTorch on ROCm, int8 on CPU)
Audiobook tab tidied up: the settings column is now grouped into compact collapsible sections (Output / Book details / Pronunciation / Markup), so script + voice + Create sit up top instead of a long scroll — same controls, denser layout (#1214)

Removed

The Dubbing per-segment picker's hardcoded design-presets group — superseded by the richer designed-voice Gallery; already-saved preset: picks still generate identically (#1220)

Added

Voice picker: the designed-voice Gallery is now selectable anywhere a voice is chosen — the audiobook default voice and each Cast row can pick a gallery archetype (searchable, favourites first), and it's materialised into a real profile on pick so it just works everywhere (#1219)
The Stories editor and the Dubbing per-segment voice pickers now use the same gallery-enabled picker, so designed-voice archetypes are selectable there too; the dub picker drops its redundant hardcoded presets group in favour of the richer Gallery (existing picks unchanged) (#1220)
Audiobook tab: a Cast panel maps each [voice:NAME] in the script to a profile so multi-voice renders correctly (it previously fell back to a single voice), plus a markup insert toolbar, live stats (chapters · words · est. runtime), and pre-flight validation for unknown voices and empty chapters (#1217)
Audiobook tab: a Stop button that truly cancels a running generation (not just the UI) and live per-chapter progress — a bar, elapsed + ETA, and each chapter's status (rendering / done / cached / failed); finished chapters stay cached so Create again resumes. Cmd/Ctrl+Enter starts a render (#1216)
Settings → Permissions + wizard System Check: live mic/Accessibility grant state, per-OS guidance, Open Settings deep-links; dictation pre-flights the mic grant (#1175)
parakeet-mlx engine: Parakeet TDT v3 on Apple Silicon — 25 EU languages, word timestamps, ~2 GB, opt-in from Settings → Models, never auto-downloads (#1175)
First-run downloads race the direct GitHub path against the mirror and use whichever answers fastest (#1179)
First-run consent question for the existing opt-in analytics (two equal buttons, skip = no)
First run: when the app auto-opens in a non-English system language, a one-time, dismissible banner offers to switch the UI to English — shown only until you pick a language, never for English systems (#1215)
Source builds carry the publishable analytics token and get the same first-run consent ask as installers; opt-in events now note the install channel (installer / docker / source) — thanks agudmund! (#1193)
Official Google Colab notebook (notebooks/OmniVoice_Studio_Colab.ipynb) — full app + API feature tour on a free T4
ROCm Docker image ghcr.io/debpalash/omnivoice-studio:rocm (+ :stable-rocm, :X.Y.Z-rocm) (#1165)
OMNIVOICE_TRUSTED_NETWORKS — comma-separated CIDRs exempted from the consumption auth gates (share PIN / API key / dictation WS); admin routes stay loopback-only (#1170)
Info/warn system notifications are dismissible and stay dismissed across restarts; error-level notices can't be dismissed, and the unclean-shutdown notice is now acknowledged server-side — thanks agudmund! (#1192)
clone_voice MCP tool — AI agents can clone a new voice from a base64 reference audio sample; returns a profile_id immediately usable with generate_speech — thanks paoloantinori! (#1194)
Dub tab: Paste Translation — paste a translation made elsewhere (ChatGPT, DeepL, a human) as subtitles, numbered lines, or plain lines; it maps onto the existing segments with a before→after preview, keeping timings and the source transcript intact (#1203)
Audiobook tab: Production Overrides (position/class temperature, steps, guidance, postprocess, seed) for expressive narration, plus IndexTTS2 emotion controls and a "vary repeated lines" toggle — defaults reproduce today's renders exactly (#1208)
Audiobook tab: a Load sample button that fills the editor with a demo story — chapters, per-character [voice:], [pause], [slow]/[fast]/[emphasis]/[spell], and reaction tags — so first-timers can hit Create and hear every capability before their real work (#1214)

CI

The quiet changelog style and 21-locale key/placeholder parity are now enforced by plain pytest checks; CodeRabbit/Greptile carry the house rules via .coderabbit.yaml/greptile.json (#1198)

Docs

docs/expressive-speech.md: per-engine breaths/laughter/emotion control, incl. the default engine's 13 native reaction tags
Flush caches / Unload documented in the performance guide, incl. POST /system/flush-memory for scripts
README FAQ: why a longer reference clip doesn't clone better (zero-shot 15 s cap; fine-tuning is the audiobook-grade path)
docs/expressive-speech.md corrected so every recipe it names (breaths, temperature) is reachable in the surface it points to, including the Audiobook tab (#1208)
New docs/api-auth.md — one place for authenticating the local API: share PIN, API key, dictation WebSocket, and trusted networks, with curl/SDK examples and what 401/403/429 mean (#1212)

Fixed

Downloading a finished audiobook (or story mix) no longer hijacks the app: in the desktop WebView a plain download link to the media file made WebKit navigate the whole window to it and play it fullscreen (then the blank-window guard misfired) — downloads now go through the native Save dialog + a server-side copy instead (#1218)
The blank-window guard's fallback page is shown by injection rather than a data: URL the desktop WebViews refuse to navigate to, and its Reload button now returns to the app even if the window had navigated away (#1218)
Security (server mode): the admin routes (/system/*, /api/settings/* — RCE-class) now require the API key or genuine loopback — with an API key set and OMNIVOICE_TRUSTED_NETWORKS configured, a trusted-network client could previously reach them with no credential; the short share PIN no longer gates admin either (#1213)
A render error no longer blanks the whole window — a recoverable error card (Reload / Report) appears instead, and CI now builds the real production bundle so a pre-mount crash can't ship (#1209)
Voice-clone trimmer: the preview now plays exactly the selected region on variable-bitrate clips (it had drifted off on VBR/mis-reported-duration files by playing the original file on a different timeline) (#1210)
Screen readers now announce the hidden file-picker buttons (batch add, gallery import, stories import) (#1211)
Audiobook language selection now reaches the backend — the client had dropped the language field, and the tab's Markup reference now lists the reaction tags ([laughter], [sigh], …) that already work there (#1208)
A backend that fails to start now says why — exit code and error output, with actionable hints and a one-click report — instead of the evidence-free "Can't reach the local OmniVoice backend" (#1177)
Generation no longer crawls on CPU after a cancelled or failed dub: the TTS model is moved back to the GPU on every exit path, and each generation now verifies its own placement (#1191)
A generation queued behind a busy one no longer spends its timeout waiting: the budget starts when a GPU worker picks the job up, so a queued request can't be failed as "too heavy for the available compute" without having run (#1190)
One request's timeout no longer cancels unrelated jobs already waiting in the GPU queue (#1190)
Timeout messages stopped claiming capacity was restored automatically — the abandoned job keeps the device until it finishes, and the guidance now says to let it drain (#1190)
The length-scaled generate budget now covers every path — streaming previews, batch dubbing, /v1/audio/speech, dub and archetype previews — instead of only the two classic call sites, so long inputs stop failing at a flat 300s (#1190)
Provenance watermarking moved off the GPU worker pool: on 1-worker machines each embed was serializing ahead of the next generation (#1190)
A batch segment that times out fails the job with a reason instead of shipping a finished-looking dub with silent gaps (#1190)
/v1/audio/speech refuses work up front with 429 + Retry-After when the pool is saturated, and returns a retryable 503 rather than a 500 on timeout (#1190)
Subtitle parsing no longer stalls on a blank-line-heavy .srt: the timing-line regex backtracked across newlines, so a mis-saved export could pin an import for hours (#1203)
A broken ASR engine's fallback could silently auto-download multi-GB weights — every fallback now passes the same no-download preflight and shows the download CTA instead (#1189)
Dub transcription releases the ASR model from VRAM on every exit — crashes, early errors, and client disconnects included (#1175)
An invalid dictation model override could bypass the missing-model check for the Whisper fallback (#1175)
The OpenAI-compatible transcription route's 409 now carries the same typed download-CTA payload as every other route (#1175)
Cross-drive installs with a user-pinned UV_CACHE_DIR still keep uv's managed Python off the system drive (#1189)
MCP clone_voice accepts data-URI base64, stores the clip under its real container extension, and returns backend validation errors as structured JSON (#1195)
Sidecar launch errors no longer embed the user's home directory in logs or error messages (#1189)
KittenTTS degrades gracefully if a future upstream update moves its text chunker (#1189)
Restored 151 broken locale strings: mangled {{placeholders}} (vi/ar showed literal _V_0__) and gallery errors that dropped their detail in all 20 translations (#1198)
Voice cloning no longer fails on quiet recordings — silence removal retries with gentler thresholds (then skips), and a truly silent clip gets a localized, actionable message instead of a dead-end 400 (#1188)
Custom install folders on another drive are honored end-to-end — uv's wheel cache and Python download now follow the chosen environment folder instead of filling C: (#1186)
Dubbing no longer fails outright when an ASR engine's dependencies are broken ("No module named 'lightning_fabric'") — the engine is marked unavailable with a repair hint and the next one is used automatically (#1185)
/v1/audio/speech no longer 500s with a raw "Exec format error" when bin/ holds a zero-byte GGUF placeholder — managed binaries are validated before exec and broken ones return an actionable 400/503 naming the repair (#1172)
KittenTTS no longer aborts with "invalid expand shape" on digit-heavy input and no longer 500s on empty input — chunks are split to the ONNX 512-token cap and unspeakable text returns a clear 400 (#1173)
Quitting the app while a model is still loading now shuts down clean — no "cannot schedule new futures" traceback, no crash-shaped exit or phantom crash record, and first-run/upgrade boots keep their backend log (#1174)
Dictation falls back to the main ASR engine when a model transcribes real speech to nothing (sherpa NeMo-TDT decoder defect, upstream k2-fsa/sherpa-onnx#3767), remembers the demotion, and names the model that failed (#1175)
A failed dictation session with no transcript clears the floating pill instead of parking it on screen forever (#1175)
A TTS-only first run gets a one-click ASR download prompt instead of a silent 1.6–3 GB Whisper pull (dub, batch, dictation, clone-ref, /v1 STT, boot warm-up) (#1175)
App boot makes no Hugging Face calls and never silently downloads the TTS model — warm-up is local-cache-only (#1175)
Quitting with a batch dub in flight no longer hangs shutdown (#1175)
Dubbing's vocal-separation step no longer crashes on Windows dev runs (SelectorEventLoop sync-pipe fallback) (#1184)
Closing the app while the model is loading is logged as a shutdown, not a phantom "Model loading failed" crash (#1183)
The backend-crash banner sits below the navbar and stays clickable (#1182)
Windows source builds: bun desktop no longer dies on a UnicodeEncodeError from piped status glyphs (#1181)
The Windows app no longer boots to a black screen (production-minifier variable reorder in the splash; now gated by a real minified-bundle e2e) (#1178)
No more storms of black console windows on Windows — every subprocess, incl. third-party spawns, runs windowless (#1178)
bun desktop self-heals a stale terminal PATH without ~/.cargo/bin and hints when Rust is genuinely missing (#1180)
FunASR/SenseVoice: no more crash or speaker-identity swaps across 30 s chunks with inline diarization (#182)
Provenance watermark now applied on every synthetic-audio route via one chokepoint (#1169)
"Can't reach the local backend" reports unclean backend deaths with evidence in dev/Docker/LAN too (#1164)
"Setup failed" screen renders instead of crashing — thanks bultodepapas! (#1159)
Backend error logs keep stack traces (swept across 20 sites) — thanks bultodepapas! (#1160)
Reference-clip uploads can't hang on a stuck audio probe (10 s timeout) — thanks bultodepapas! (#1162)
Malformed EPUB chapters import partially instead of vanishing — thanks bultodepapas! (#1161)
Failed sidebar fetches are logged instead of silently showing stale/empty lists — thanks bultodepapas! (#1158)
A missing/broken mcp package degrades to "/mcp disabled" instead of killing the backend at startup (#1156)
"Setup failed" auto-dismisses when the backend recovers; relaunching retries instead of refocusing a dead window (#1156)
Ended the forrtl: error (200) mid-session Windows crashes (math-runtime console handler) (#1153)
Non-Latin text can't crash synthesis on Windows (backend forced to UTF-8) (#1155)
Video-export errors diagnose the real cause (Windows 32k command-line limit; big filter graphs go via a script file) with per-mode advice (#1152)
Remote backends with an API key get an API-key prompt (durable per-browser, one-shot #api_key= link) instead of an unpassable PIN form — thanks paoloantinori! (#1154)

macOS Apple Silicon artifacts

36aae1bf0b16772c6c4f8ff1be06c3368b5d28b4f424ba133e554d7a062ac76b  OmniVoice Studio_0.4.0_aarch64.dmg
1868ca47bf4a9659d2b36b658b96e7febdb8a419b2802b918752ae452066ddca  OmniVoice Studio.app.tar.gz
e157fcbf06af0c81847430bf76ff158cf64f39c23c240707df106ccf80657331  OmniVoice Studio.app.tar.gz.sig

macOS Intel artifacts

e194dcdab30f913a08ff9487e96d5f550e50dc2c5917928768a896d3a494f799  OmniVoice Studio_0.4.0_x64.dmg
5272f36a7b657b1e6d68b51e379cd2332abb8274eb85db857716f199f74d91ec  OmniVoice Studio.app.tar.gz
b2e3e5e30fb2dd1bc1a1fb76f3cead0e55edfea755d9bc9b451f0f05484273ee  OmniVoice Studio.app.tar.gz.sig

Windows x64 artifacts

0d7358c2ad48a6a84cbf18af89a3a33aa835fff211e82a96b54f5cbcdd85d729 *OmniVoice Studio_0.4.0_x64_en-US.msi
d67bf947f0fad7038b3a3d3e1eef1f9ba2fdcb48ca10aed84cd7f33d427da9de *OmniVoice Studio_0.4.0_x64_en-US.msi.sig

Contributors

Thank you all 💜

详细ChangeLog

v0.3.22

2026年07月14日

The dubbing release. Dubbed videos stop sounding like a compromise: the music keeps its stereo width and full frequency range, short lines no longer leave dead air while the mouth keeps moving, one speaker stays one voice, and the language tabs finally switch the transcript with the audio. Underneath it, the memory fixes that ended the "can't reach the local backend" era on 16 GB machines ship at last — plus a sweep of never-again hardening drawn from an audit of every bug this project has ever closed.

Added

A "Voice match" toggle for dubbing — keep one steady voice per speaker. Each dubbed line clones from a snippet of its own original audio, which matches the delivery beautifully but can make the voice itself drift from line to line — most audibly on videos where speaker detection ran in fallback mode ("still 4 segments different in voice", as one report put it). A new control next to the Timing picker chooses: Per line (the default, unchanged) for the best per-line delivery match, or Consistent to clone every line of a speaker from one shared reference — the speaker's pooled sample, or the best single clip when none exists — for a steady identity across the whole dub. Flipping it honestly marks segments as needing regeneration, and the shared reference is encoded once and reused, not re-studied per line. (#1147)
A performance guide, at last. docs/performance.md explains where generation and dubbing time actually goes, the three classic causes of "it got slow" (an empty Transcript field on a voice profile chief among them), every tuning knob the backend reads — none of which were documented anywhere — and which settings to leave alone (raising OMNIVOICE_GPU_WORKERS on a small GPU is how you get the crash the default exists to prevent). Includes how to run the built-in profiler so a slowness report can carry numbers instead of vibes.
In-app analytics is now wired end to end — and still off until you say yes. The frontend analytics SDK is only ever started after you opt in (Settings → Privacy), never at app launch, so a default install still transmits nothing. Two of the SDK's defaults are explicitly disabled because they would be actively harmful here: autocapture, which sends the text content of whatever you click — in this app, the script you are about to synthesise, your voice names, your file names — and session recording, which records the screen. Events carry metadata only, filtered through the same allowlist as the backend, so no future change can leak your content by adding a field.
Opt-in analytics — off by default, and it can't lie to you. OmniVoice still sends nothing out of the box: no accounts, no telemetry, no phone-home, and your text, audio, voices, and projects never leave your machine regardless of what you choose. There is now one toggle in Settings → Privacy → "Help improve OmniVoice", off unless you turn it on. If you do, it sends anonymous usage stats — which engine and language you used, how long a generation took, how many characters the text had (a number, not the text), and the type of any error. It never sends the text you type, your audio, your file names, your voice names, or anything identifying you. That isn't a promise in a policy: an allowlist in the code drops any property that isn't on it, so a future change can't leak content by accident, and crash tracebacks are deliberately not auto-captured (they can carry file paths and tokens). Turning it off stops everything immediately. Builds from source have no analytics destination at all and don't even show the toggle.
Settings → Usage: see what you've made, counted entirely on your own machine. Takes generated, audio produced, voices, days used, and a breakdown by mode and language — all computed from the history already in your own database. It collects nothing new, stores nothing new, and transmits nothing anywhere, no matter what you've chosen under Settings → Privacy: this panel is yours, it works with analytics switched off, and it never phones home. If you want to know what you've been making, the answer shouldn't require sending it to anyone.
The memory panel now tells the whole truth. Settings → Models (and GET /model/loaded) used to report only the OmniVoice core model — a resident second engine like MLX-Audio, or the warm dictation model, was invisible, so the memory picture looked ~2 GB lighter than reality. It now lists every resident model (in-process engines and the dictation ASR included) and adds a system block with free/total RAM (and free VRAM on a dedicated GPU) plus a low-memory warning. On top of that, a load that starts while memory is already low leaves a breadcrumb in the backend log, so a subsequent out-of-memory kill points at the load that tipped it instead of dying silently. Advisory only — nothing is blocked (the OS can reclaim memory, and refusing a load on an estimate would brick machines that would actually cope). Tune the threshold with OMNIVOICE_LOW_MEMORY_HEADROOM_GB (default 2).

Fixed

Switching preview languages can't leave a mixed-language transcript. Follow-up to the tab/transcript sync: if a track's translations were only partially stored in the browser (older projects, partial regenerations), switching tabs could show German audio with a few rows still in the previous language. Missing rows now hydrate from the app's own per-language store on the backend — and a picked regional dialect is automatically cleared when you switch to a language it doesn't belong to, wherever the switch comes from. (#1149)
The Export step's language tabs now switch the transcript too. Clicking Bengali/German/Hindi… above the finished dub swapped the video but left the segment list showing whichever language you generated last — German audio over Bengali text. The tabs now also swap every segment's text to that language (through the same per-language store the language picker uses, so nothing is lost when you switch back); the Original tab keeps your editing language as-is, since each row already shows the original line beneath its translation. (#1148)
A "backend crashed" notice can no longer outlive the update that fixed the crash — and the desktop shell's self-repair paths are now pinned by tests that CI actually runs. Crash notices now record which app version wrote them, and a notice left behind by an older version is ignored and cleaned up after you upgrade instead of resurfacing as if the new build had crashed. The Windows blank-window repair (the one-click WebView cache fix after a BSOD) also gets regression tests pinning its safety contract — one attempt per request, never touches anything unasked, never blocks startup on a locked cache — and CI now runs the desktop shell's entire Rust unit-test suite on macOS, Windows, and Linux, which it previously never executed at all. (#1145)
The MLX-Audio phonemizer's language model now ships with the app environment instead of being fetched mid-generation. Follow-up to the pip fix: with the installer present, the first English MLX-Audio generation would auto-download a small model straight from GitHub — an outbound request that bypasses the app's mirror system (a problem on restricted networks) and fails offline. The model is now a pinned dependency of the managed environment: it arrives at install/update time through the normal dependency flow, and first generation works fully offline. (#1146)
The MLX-Audio engine's first English generation no longer trips over a missing installer. Its phonemizer auto-downloads a small language model on first use by shelling out to pip — which the app's managed Python environment didn't include, so the download always failed (and before the recent containment fix, took the whole backend down with it, #1133). pip now ships as a real dependency of the managed environment, so it survives app updates too — anything installed ad-hoc would have been stripped by the updater's environment sync, quietly re-breaking this after every release. (#1144)
A voice engine's helper library can no longer shut down the whole backend. One user's backend died 21 seconds after starting (#1133): the MLX-Audio engine's phonemizer tries to auto-download a language model on first use, the downloader is written as a command-line tool, and on failure it calls "exit the program" — which, running inside the backend, exited the backend. Any engine dependency written that way could do this. Exits are now contained at the engine-dispatch boundary and turned into a normal, explained error ("an engine dependency failed to auto-install something — see the log"), for TTS and transcription alike. The app keeps running; the failed request tells you what actually happened. (#1143)
Vietnamese years read like Vietnamese again. A recent release started spelling out numbers before synthesis, and its Vietnamese number library turns out to be wrong for exactly the numbers people say most — years ("2024" became "hai nghìn lẻ hai mươi bốn", which no Vietnamese speaker says). The voice model has always pronounced Vietnamese digits correctly on its own, so Vietnamese text now keeps its digits — the same conservative rule that already protected Vietnamese decimals. Also closes the loophole that made this depend on spelling: picking "Vietnamese" from the language list behaved differently from the code "vi". (#1139)
A voice profile's pinned seed now pins Audiobook renders too. Locking a take (or a designed voice) stores a seed so the voice performs reproducibly — and the Voice page honors it, but Audiobook/Stories renders quietly ignored it and rolled fresh randomness for every segment. Book renders with a pinned-seed profile are now deterministic end to end, matching the Voice page. And the audiobook renderer's higher generation quality (32 decoding steps — the model's own quality preset, vs. the Voice page's fast default of 16) is now pinned explicitly in code rather than inherited by accident, so it can't silently change; that steps gap is also why Audiobook sounds steadier than Voice at default settings — move the Voice page's Steps slider to 32 for the same quality. (#1139)
A finished audiobook's Download button stops vanishing. The player and Download link for a completed book lived only in the page's temporary state — switch tabs once and they were gone, which read as "no way to export at all" (the file was still on disk, and in Projects → Audiobooks). The last finished render now survives tab switches and reloads, right where the book was made. (#1139)
Six recurrence guards from a full audit of the project's issue history — aimed at "this bug can never come back, even after an update or reinstall." (1) Before loading the voice model on a memory-tight machine, the app now first releases things it already reclaims on idle (the warm dictation model, allocator caches) — the missing half of the 16 GB OOM-kill fix; roomy machines pay nothing. (2) When the operating system force-kills the backend for running out of RAM, the crash notice now says exactly that instead of blaming "VRAM" on machines that have none. (3) Saving a cloned voice with free-form text in its delivery field can no longer persist a profile that errors on every future generation — the server now sanitizes all profile kinds, closing a hole that had been re-exploited three times through different clients. (4) A reinstall that inherits an old settings file pointing at an unplugged drive or deleted folder no longer sends downloads into the void — dead paths are ignored for the run with a clear log line. (5) Locally-saved UI state is now schema-checked as a whole on restore, so one corrupted field can't silently discard everything after it (the general form of the "app got empty" fix). (6) File moves across drives (Windows D:-drive installs) get a dedicated safe-move helper, so the next code path that renames across devices degrades gracefully instead of failing with [Errno 18]. Long texts also get a generation time budget that scales with their length instead of a fixed five minutes. (#1141)
Dubbed videos get their stereo back — and the music's full frequency range. A/B-measuring a dub against its original showed the dubbed audio was mono in a stereo container (channel correlation 1.000 vs the original's 0.754) — the entire stereo image of the music, gone. Two causes, both fixed: the separation step was being fed the 16 kHz mono file extracted for transcription — so the music bed inherited mono and an 8 kHz ceiling at the source — and the mixer then let the mono voice drag the whole mix down to mono. Ingest now makes a second, full-quality stereo extraction (44.1 kHz) just for separation, transcription keeps its mono file, and the mixer pins both sides to stereo with the voice dead-center where dubbed dialogue belongs. Loudness already matched the original (−17.2 vs −17.8 LUFS, measured); now the width and brightness do too. (#1138)
Dubbed lines that finish early no longer leave dead air — they now speak at the pace of the scene. Translations routinely come out shorter than the original delivery, and the dub used to just stop early: measured on a real dub, 8.8 of 18.7 seconds of speech time had no voice at all — the mouth kept moving on screen over the thin residue the vocal separation leaves behind, which reads as silence and as "the music got quiet". Short lines are now gently slowed toward their time slot (pitch preserved, never below 0.85× — comfortably natural), so speech covers the speaking time the way the original did. This also does most of the work people expect from "lip sync": the voice now starts and ends with the mouth. Near-full lines are left untouched, the per-segment badge shows the applied rate, and OMNIVOICE_UNDERRUN_MIN_RATE=1.0 turns the fill off. (#1137)
The dub's background music no longer comes out quiet and muffled. Every dub export mixes your synthesized voice over the video's separated music/ambience bed — and that mix had two fidelity bugs stacked on top of each other. The mixer normalizes its inputs, so the weights meant to gently favor dialogue actually played the music at ~57% of its original level (measured); and because the voice track is synthesized at 24 kHz, the mixer silently pulled the 44.1 kHz music down to 24 kHz — deleting everything above 12 kHz: cymbals, brightness, air. The batch pipeline was harsher still, pinning the bed near 8%. All six mix sites now share one filter that resamples both sides up to 48 kHz, cancels the normalization so the music plays at 90% of its true level (a hair of headroom keeps dialogue legible), and adds a transparent peak limiter. Measured on a real dub: bed level 57% → 90%, bandwidth 12 kHz → 24 kHz. (#1136)
A rate-limited translation polish pass no longer sabotages the dub — or lies about it. The Cinematic quality mode runs an optional critique-and-rewrite pass after translating. When that pass hit a rate limit (free-tier LLM endpoints throttle hard), three bad things happened at once: the app reported "N/N segment(s) failed" in red over a translate that had actually succeeded; the affected segments were silently skipped by the speech-rate fit pass and duration planner — so overlong lines went to synthesis unfitted and came out audibly time-compressed; and the two-second "retry shortly" hint the provider sent was ignored. All three are fixed: a rate-limited call now waits out the provider's own Retry-After (bounded, once) and usually just succeeds; a segment that still misses the polish keeps its plain translation, stays in every downstream fitting pass, and is reported honestly — "translated, polish skipped" as a warning with the reason, not a failure. Rows that really failed still say so. (#1135)
Dubbing kept re-studying the same speaker's voice, hundreds of times per video. Each dubbed line clones from a clip of its own source audio (that's what makes deliveries match), and lines too short to clone from fall back to a per-speaker sample. But the app's memory for already-studied voices only holds 8 — and a long dub streams hundreds of one-shot per-line clips through it, each pushing out the per-speaker samples that every other line needs. Result: the speaker sample was re-studied (~0.4 s, measured) over and over. One-shot clips are now studied without displacing anything, so the per-speaker samples stay warm for the whole dub. Nothing about the audio changes — same clips, same voices, less repeated work. (#1132)
Clicking "Install" on an engine right after opening Settings could silently do nothing. When the Engines page opens, it quietly checks each installable engine for an in-flight install to re-attach to. If you clicked Install while that check was still running, your click's status update was thrown away to keep requests orderly — so no progress panel, no error, no retry, just nothing (the install itself did start in the background; the UI simply never showed it). Fast machines usually won the race, which is why this mostly showed up as a once-in-a-while CI test failure. The Install click's update can no longer be dropped — it politely waits out the startup check instead. (#1131)
Cloning re-listened to your reference clip for every chunk of text — now it listens once. Before OmniVoice can speak in a cloned voice it has to encode the reference clip you gave it. That encode was being redone on every single piece of the job: long text is split into chunks, and each chunk re-encoded the same reference from scratch; so did each [pause] span, and each chapter segment of an audiobook. A cache to prevent exactly this was written a while back — and then quietly bypassed on the path the Generate button actually takes, so for several releases it only ever helped the API. It's now wired into every path. Measured on an M2, one encode costs 0.4 seconds, so this gives back roughly 3–4 seconds on a long paragraph and about a minute on a 166-segment audiobook — the same voice, the same audio out, just without listening to your reference clip 166 times. As a bonus, preprocess_prompt on the OpenAI-compatible endpoint now actually does something; it was being accepted and silently discarded. (#1130)
Dubbing loaded the 3 GB voice model, threw it away, and loaded it again. Before transcribing, a dub pulled the entire voice model into memory to read a single setting off it — one that is empty unless you've turned on an off-by-default flag. So it loaded ~3 GB, found nothing, released it a moment later (on Apple Silicon that's a full unload), and then had to load the very same model again from cold when it was time to actually speak. Every dub paid for that round trip — roughly 8 seconds, plus the memory churn on exactly the 16 GB machines where memory pressure is the problem. It now only loads the model when there's genuinely something to read. (#1130)
The backend stopped holding the voice model hostage while it loads the transcription model — the 16 GB dub crash. Before transcribing a dub, OmniVoice makes room by setting the TTS model aside. On an NVIDIA GPU it did. On Apple Silicon it did nothing at all — the code bailed out with "unified memory doesn't benefit from offloading". That was half right and wholly wrong: on unified memory, moving a model to "CPU" frees nothing (it's the same RAM), but the answer is to release it, not to skip the step. So a 16 GB Mac went into a dub holding the ~3 GB voice model, then loaded a ~3 GB transcription model on top of it — measured here: 4.1 GB free before, and large-v3 needs 3 — and the operating system killed the backend mid-transcription. That's the dub that "dropped before emitting any segments". The voice model is now genuinely released when memory is tight (and left alone when it isn't, so a roomy machine pays nothing); it reloads by itself on your next generation. (#1119)
Dubbing on a Mac was transcribing on the CPU — with the GPU sitting idle. OmniVoice picked its transcription engine without ever looking at your hardware: WhisperX won every time, and WhisperX (like faster-whisper) is built on CTranslate2, which has no Metal backend at all. So on Apple Silicon it ran whisper-large-v3 on the processor. Measured on an M2, one 30-second chunk: 90 seconds on the CPU versus 20 on the GPU — slower than realtime, which turned a 16-minute video into a ~48-minute transcribe that looked exactly like a hang. Worse, the slowest chunks blew past the 2-minute per-chunk timeout and were abandoned entirely, so the transcript came back with pieces missing and the app blamed a "VRAM-starved GPU" — on a machine that has no VRAM. Apple Silicon now uses MLX, which runs the same whisper-large-v3 on the GPU, roughly 4x faster. Word timing is unchanged: the wav2vec2 forced alignment that lip-sync depends on (±10-30 ms, versus Whisper's own ±100-300 ms) is layered on top exactly as before. Same model, same alignment, four times the speed. Nothing changes on NVIDIA or Linux, where WhisperX already used the GPU. (#1127)
The transcribe screen invented its ETA, and the number was a fiction. It assumed transcription runs at ~20x realtime — true on a fast GPU — and predicted from the video's length alone. For a 16-minute video it promised 56 seconds. Once reality overran the guess it pinned itself at "~0s remaining" with the bar frozen at 95%, and sat there for the next three quarters of an hour. It now reports the real fraction of the audio transcribed and extrapolates the time left from the speed it can actually observe — so it is right on a fast machine and a slow one, and says nothing at all until it has something true to say. (#1127)
Analytics you switched on would have stayed half-dead. The backend half of the new opt-in analytics read its destination from an environment variable that nothing on your machine ever set — so in a shipped build it could never send anything, silently, no matter what you chose. Only the frontend half worked. The destination is now baked into the desktop shell at build time and handed to the backend when it starts, so "on" means on. Nothing else changes: it stays off until you opt in, builds from source still have no destination at all, and the property allowlist still decides what may leave. (#1123)
A dub that dies mid-transcription still guessed at the cause. v0.3.20 taught it to check the crash report before blaming the ASR model — but it checked instantly, the moment the stream dropped, and the desktop shell needs about two seconds to notice the backend died and write that report. So it kept looking too early, finding nothing, and falling back to the same old guess ("Likely ASR backend failed to load") even when the backend had in fact just crashed. It now waits for the shell to catch up, so you get the real cause — exit code and error output — instead of a guess. (#1119)

Windows x64 artifacts

1c14a49a18efe60e487600b10ec29e15b4cd5ffc65709b2d19324c16cef65cc3 *OmniVoice Studio_0.3.22_x64_en-US.msi
8976ad6c7767bd342882be3ce750dd27bf62db5391befe9cda5a7aaf57870473 *OmniVoice Studio_0.3.22_x64_en-US.msi.sig

Linux x64 artifacts

a375e5d9461c255ca1e02400ae9c2130505ed636f2cfc940c7ffb724718d487c  OmniVoice Studio_0.3.22_amd64.AppImage
4bc45200bf41a0b15ed6497d771e5c1bbf85517ebadb6e3f487d9c297e2719c5  OmniVoice Studio_0.3.22_amd64.AppImage.sig

macOS Intel artifacts

010e88ed715b5a75330e2fb09e2e6a5e7ed71d642fef960c4db64db067399fb9  OmniVoice Studio_0.3.22_x64.dmg
84b804c4c34ffe863b934d85a5045243c23c22b3de019871c620866fb5bd1dca  OmniVoice Studio.app.tar.gz
f9b101a4cdf77d6f9aaabcae6e331d5ba1eb58a1e777f166fd66223318078283  OmniVoice Studio.app.tar.gz.sig

详细ChangeLog

v0.3.21

2026年07月12日

The memory release. The reason the app kept saying "Can't reach the local backend" on 16 GB machines was never really the network — the backend was quietly running out of memory and getting killed. This release fixes that at the source: the models it holds now get out of each other's way. Plus the uninstaller and factory reset grew into a proper Settings → Storage pair.

Added

Factory reset grew up: Settings → Storage → "Reset & remove". It used to do exactly one thing — clear your UI preferences — while the only other option was deleting everything and starting over. Between "forget my theme" and "wipe the machine" sat every reset people actually needed. Now there are four one-click tiers — UI preferences, all settings, downloaded assets & models, and everything OmniVoice did — plus a per-item checklist if you want to drop just the model weights, just a wedged sidecar engine, or just the history. Every option shows its real size on disk before you commit, and the number on the button is exactly what gets freed. Deleting voices, projects or audio asks you to type DELETE; nothing irreversible happens on a single click. "Everything" deliberately stops short of the Python environment, so you land on a working first-run screen rather than a rebuild — the app stops its engine, deletes, and starts it again for you. On macOS and Linux the model cache is the shared Hugging Face cache, so it's its own checkbox and says so; on Windows and portable installs it's OmniVoice's own, and the app doesn't pretend otherwise.
The Storage panels got a design. "Remove all data" and "Reset & remove" listed folders as a flat run of text, so a 7.5 GB model cache and a 391-byte config file carried exactly the same visual weight — the one thing you actually wanted to see (where the space went) was the one thing you couldn't. Every row now has an icon, a dimmed path, and a proportional bar showing its share of what will be freed, so the big one looks big. The shared Hugging Face cache is promoted out of the confirm dialog into its own "Optional" row with a checkbox, so ticking it moves the running total in front of you instead of springing a different number on you at the point of no return, and the dialog now lists exactly what is about to go.

Fixed

Switching TTS engines no longer stacks their models in memory. Using a second engine in a session (or a per-request engine override) loaded its model on top of the first one's, because the OmniVoice core model and the other engines live in two separate caches that never coordinated — measured on a 16 GB M2, an omnivoice → mlx-audio switch left the machine holding both (footprint 3.9 GB → 4.3 GB, the ~2.8 GB core never freed). That accumulation is a direct contributor to the memory pressure behind the "Can't reach the local backend" OOM deaths. Now only one TTS engine's model stays resident: resolving an engine hands back every other resident engine first (the same omnivoice → mlx-audio switch now drops to ~1.5 GB). Steady-state single-engine use is unaffected; an A/B switch pays a re-load on the way back (~8 s for the OmniVoice core, ~1–2 s for the lighter engines). Opt out with OMNIVOICE_SINGLE_ENGINE_RESIDENT=0 if you have RAM to keep several warm. Two underlying leaks are fixed as part of this: every in-process TTS engine's unload() now actually frees its model and empties the device cache (previously all but OmniVoice were silent no-ops), and faster-whisper's unload() cleared the wrong attribute so its model was never released.
The backend no longer sits on ~2 GB of idle dictation model — the real reason it was being killed on 16 GB Macs. Four reports of "Can't reach the local OmniVoice backend" (#1076, #1092, #1093, #1101) all died at the same moment: during a generate, on a 16 GB machine. Measuring it showed the generate was never the problem — it costs about 116 MB. The problem was the baseline: the backend sat at ~6.2 GB even while idle. The TTS model has always been unloaded after an idle timeout, but the speech-recognition model used for dictation never was — so once you dictated a single time, ~2 GB stayed resident for as long as the app ran. On a 16 GB Mac, that plus the app, macOS, and your other programs is enough for the system to run out of memory and kill the backend, which surfaced as the "can't reach the backend" error. Dictation's model now gets the same idle release the TTS model already had, handing that memory back. The only cost is a ~1.4-second re-warm on your next dictation after a long pause, and a live dictation session is pinned so nothing is ever unloaded mid-sentence.
Folder sizes under 1 KB displayed as "0 KB". The uninstall panel's 391 B config folder rendered as 0 KB — which reads as "nothing here" for a folder that very much exists. The Storage panels now share one byte formatter that can say 391 B.
Some styling silently did nothing. A handful of components referenced CSS custom properties that were never defined (--chrome-fg-subtle, --chrome-bg-raised, --color-warning). An undefined var() makes the whole declaration invalid, so the browser drops it and the element quietly inherits — the dimmed folder paths in the Storage panels weren't dimmed at all. Fixed in those panels, and a new guard (frontend/src/test/cssTokens.test.js) fails on any bare var(--token) in JSX that isn't defined in a stylesheet or documented as runtime-injected, so a typo can't ship as invisible styling again.

Linux x64 artifacts

2b9e5ac428ad90be0662465fcff244247c9756d30223058254385a449c3ea89e  OmniVoice Studio_0.3.21_amd64.AppImage
4234981ecae54772dba76f05b7d67b823f1f79cef5882a406813400bdf7ad87a  OmniVoice Studio_0.3.21_amd64.AppImage.sig

Windows x64 artifacts

466ed828188631bc19959e1edcf753e03a892c3159699510f9917653dd8b6cd2 *OmniVoice Studio_0.3.21_x64_en-US.msi
93fa3bde043aeb4b9d7eda5f534be5dc8d8d6817f9241935d859e0e05590f40b *OmniVoice Studio_0.3.21_x64_en-US.msi.sig

macOS Intel artifacts

aa5c9e2182ec9707d0b3939278df274f4a9ddea10e87496da12e7e538b2f2855  OmniVoice Studio_0.3.21_x64.dmg
41a69a3db3c2dbdd0955ab6aa0bb63836a9358112a2f4b895fe8d65915fdf8e9  OmniVoice Studio.app.tar.gz
ebc343d107022764e230d3efe109b4977c850d25edcf283ab9c7f13603e1b833  OmniVoice Studio.app.tar.gz.sig

详细ChangeLog

v0.3.20

2026年07月12日

The follow-through release. v0.3.19 promised that "Can't reach the local OmniVoice backend" would stop firing while the backend was merely restarting — and then a user hit it anyway, on 0.3.19, because the fix had a race in it. That's closed properly here. Uninstalling also stopped being a thing only maintainers could do: it's now a button in the app, where the person who asked for it can actually reach it.

Added

Uninstall is now in the app: Settings → Storage → "Remove all data". The v0.3.19 uninstaller was a script — which never reached the people who needed it, since anyone who installed the .dmg / .msi / AppImage has no repo to run it from (exactly the case in #1089). The app now lists every folder this install owns with its real size, deletes them behind a typed confirmation, and quits. The downloaded model weights are a separate, opt-in checkbox, because that's the standard Hugging Face cache shared with other AI tools on your machine — removing it can delete models OmniVoice never downloaded. Custom and portable install locations are honored, and nothing outside OmniVoice's own folders can be touched. The scripts now also ship as release assets, so you can clean up without launching the app at all. (#1089)

Fixed

"Can't reach the local OmniVoice backend" could still fire on 0.3.19 — the fix had a hole. The app asks the desktop shell whether a start/restart is in progress before showing that error, but the shell learns of a dead backend from a 2-second poll: when the backend dies mid-generation, the supervisor needs a moment to notice it, record the crash, and flip its state to "restarting". The app was asking once, ~3 seconds in — often still hearing "everything's fine" — and dead-ending on the generic toast anyway. A failed connection contradicts "everything's fine", so that answer is now treated as stale rather than authoritative: the app keeps retrying briefly, letting the shell catch up, which turns the failure into the "backend is restarting — hang tight" banner (and gives the crash report time to be written, so you get the real cause instead of a guess). A shell that has genuinely given up, or no shell at all, still errors immediately. (#1101)
The uninstaller was leaving the backend's log folder behind on Linux and Windows. It cleaned the app-data, config, and Python-env folders but missed where the backend actually writes backend.log / backend_err.log — ~/.local/state/OmniVoice on Linux and %LOCALAPPDATA%\OmniVoice\Logs on Windows. Both the scripts and the documented path lists now cover them. (#1089)

macOS Apple Silicon artifacts

0154046993f87772b5bf9b59ec37b7c86c7165aef1d7bc1f1b739dad80f198b6  OmniVoice Studio_0.3.20_aarch64.dmg
e040ef798d57cbd53fefb32c92156be613a4fa0ba257a074d207a0376b7ebe04  OmniVoice Studio.app.tar.gz
6d88bb653fcff10bbe8ac5f52349ea77ab3a2529caca1b1d04188b64a072f1a1  OmniVoice Studio.app.tar.gz.sig

Windows x64 artifacts

6288e6d6e8b3a25bbb73fa8b3df48c710435583f30814dc741c49ef7305df7bb *OmniVoice Studio_0.3.20_x64_en-US.msi
d2cb4eea0f2f22e2e4e490419b479b7eb624107a152bf98063cdbc7d68e6ee52 *OmniVoice Studio_0.3.20_x64_en-US.msi.sig

macOS Intel artifacts

4efa7bcf171d99889fddc3b57757267bdaa34113f1560dfb80f0998e91e765b3  OmniVoice Studio_0.3.20_x64.dmg
592e4aa42ce75e62a14d359a037d8aceb89ca1dc4a2d89c4ee199e2340d6d326  OmniVoice Studio.app.tar.gz
2064e0597271f7eab2b1308efba927e2cf339ec968821a4b85ce2358bd3e1eff  OmniVoice Studio.app.tar.gz.sig

详细ChangeLog

v0.3.19

2026年07月12日

The honesty release. Every error in here was already technically true and practically useless — so this round went after the lies the app tells when something goes wrong. "Can't reach the local OmniVoice backend" no longer fires while the backend is simply still starting; a dead Hugging Face mirror no longer strands the setup wizard with advice it can't follow; and a dub that dies mid-transcription now names the actual cause instead of guessing at it. Alongside that: generated speech starts playing on the first chunk instead of the last, and there's finally a real uninstaller.

Added

Generated speech starts playing on the first chunk, instead of after the last one. Long text is synthesized in chunks, but you used to sit through the entire render before hearing anything. The Studio now streams the preview: audio begins the moment the first chunk is ready and the rest arrives as it renders, so a long passage is audible in about the time the first sentence takes. The take saved to your history is byte-identical to the non-streaming render — streaming is a delivery channel, not a different synthesis path — and if a stream fails mid-flight the app falls back to the classic whole-file flow with nothing half-written to disk. (#1088)
A clean uninstaller + a straight answer to "where's my data?" OmniVoice is fully local, so removing it is just deleting the folders it wrote — but until now users had to guess which ones. New scripts/uninstall.sh (macOS/Linux) and scripts/uninstall.ps1 (Windows) find every OmniVoice folder — app data, the multi-GB managed Python env, config, logs, and (separately, because it's shared) the Hugging Face model cache — print each with its size as a dry-run first, and delete only on --yes. They honor your custom locations (OMNIVOICE_DATA_DIR, HF_HOME, portable mode) and never touch the app binary. The complete per-platform path list lives in the new docs/install/uninstall.md, linked from the README FAQ, SUPPORT, and troubleshooting. (#1089)

Fixed

A dub that dies mid-transcription now says what actually happened instead of guessing. "Transcribe stream dropped before emitting any segments. Likely ASR backend failed to load" was a guess — and usually the wrong one. The backend is contract-bound to emit a terminal event on every stream even when it fails, so a stream that simply goes silent means the backend process died underneath it — on smaller GPUs, almost always a native out-of-memory abort while loading the ASR model on top of a still-resident TTS model. The app now consults the desktop shell's crash forensics and tells you that: the exit code, when it happened, a one-click "View crash details" with the captured error output, and the actual next step (free VRAM / pick a smaller ASR model) rather than "check the backend log". With no crash recorded, the original message still stands. (#1062)
"Can't reach the local OmniVoice backend" stopped crying wolf during startups and restarts. A real backend start or auto-restart takes 10–20+ seconds (Python spawn plus the PyTorch import), but the app's transport retry only bridged ~3 seconds — every click inside that window dead-ended with the scary toast, over and over, even though the backend healed itself moments later. The app now asks the desktop shell whether a start/restart is actually in progress and simply waits for it (up to the shell's own 2-minute restart budget), and shows a single "backend is restarting — hang tight" banner with a "back — carrying on" confirmation — the reconnecting affordance the supervisor has promised since #567. A truly dead backend (or a non-desktop deployment) still errors promptly, and the crash notice keeps telling the honest story.
A dead Hugging Face mirror can no longer strand the first-run wizard. When a model download failed because the configured mirror was unreachable, the error pointed at Settings — which first-run users can't open (the wizard gates the studio) — and falsely claimed the mirror setting only applies after a restart (downloads actually pick it up per call, immediately). Now the wizard shows the mirror quick-pick (including "Hugging Face (official)") right next to the failed download and retries it the moment you switch; the corrected hint says retry-first, restart only if it still fails. Two backend holes in the same flow are closed too: switching endpoints clears the "failed recently" retry cooldown (no more 429 on the immediate retry), and clearing to official also removes the legacy hf_endpoint pref, which used to silently keep the dead mirror in effect.

Changed

The first-run wizard shows the app version in its masthead, next to the OmniVoice Studio title — so setup-time screenshots and bug reports identify the build at a glance (the install splash already did).
Repo root decluttered. Retired the finished planning archives (.planning/, specs/), the pre-React design mockups (design/), the legacy research dir (research/), and stale third-party agent rules (.agents/) — ~110 files of process noise gone; everything stays in git history, and the four load-bearing engine decision docs moved to docs/adr/. Contributor-facing only; the app is unchanged.

Linux x64 artifacts

4aae088250f48d6da432ea10483875ccf26647b60be67de65cfebf465cc6622b  OmniVoice Studio_0.3.19_amd64.AppImage
0fada05ba551b7a2f325fd9ad132f1c48dcf13ee8c6e005120410443a2eb218c  OmniVoice Studio_0.3.19_amd64.AppImage.sig

Windows x64 artifacts

373a7eace6dd9ae279f5ccadfe681dd5beb1b1144981a5ca38cf44e4d1954ef0 *OmniVoice Studio_0.3.19_x64_en-US.msi
3b2083d13093031137e46f46693b5851b01d13cdf9c111d7eecf1aae32ca5c6f *OmniVoice Studio_0.3.19_x64_en-US.msi.sig

macOS Intel artifacts

70499c162e027395894b8a4f0c959e69a926f1a660036a2c552fe106dd69e789  OmniVoice Studio_0.3.19_x64.dmg
586bdb50baf290476bfdc011d6f1a8f2355d919d8aff8b342027fdebe2ff8d93  OmniVoice Studio.app.tar.gz
570c608c81e67ca975f6a90ec07ad3c2336e1b0daa71e1393149944bff278f92  OmniVoice Studio.app.tar.gz.sig

详细ChangeLog

v0.3.18

2026年07月12日

The self-sufficiency release. Two long-standing "works on my network / works after four terminal commands" walls came down: model downloads now find a reachable Hugging Face endpoint on their own (no more restricted-network first-run dead-ends), and IndexTTS-2 — previously the only engine that demanded a manual clone-venv-install ritual — installs itself with one click. Under the hood, a test-debt sweep hardened the suite that guards all of it.

Added

IndexTTS-2 installs itself now — one click in Settings → Engines. The emotion-controlled cloning engine used to demand four terminal steps (clone the repo, create a venv, uv pip install, set an environment variable); the row now has an Install button that does all of it — source fetch (git, with a no-git tarball fallback), an isolated venv that keeps its transformers<5 away from the app, the ~6 GB model weights (via your configured/auto-selected Hugging Face endpoint), and configuration — with step-by-step progress, a disk-space check before anything is written, and resumable repair if anything is interrupted. The engine is usable the moment the job finishes, no restart; existing manual installs are detected and left untouched, and the manual steps remain as a collapsible fallback. The provisioner is parametrized so future sidecar engines (MOSS-v1.5, dots.tts, Confucius4) can reuse it. (#1083)
Model downloads now find a reachable Hugging Face endpoint on their own. On networks where huggingface.co is blocked or slow (the class of first-run dead-ends behind #984), the app quietly probes the official endpoint and the hf-mirror.com community mirror, picks whichever actually works, remembers the choice, and re-checks only when a download fails or the pick goes stale — so a restricted-network first run reaches a working voice instead of a wall of connection errors. Anyone who already set a mirror (env var, pref, or Settings) stays exactly where they pointed: explicit choices are never auto-switched, and Settings → Models → Hugging Face mirror now shows the automatic pick with its measured latency plus a "Test again" button. Probes only touch the two download hosts — no geo-IP, no telemetry — and every download stays checksum-verified by huggingface_hub regardless of endpoint. (#1082)

Fixed

Concurrent settings writes can no longer drop each other. Two parts of the app saving preferences at the same moment (say, an engine install finishing while you change a setting) could silently lose whichever save landed first; preference writes are now serialized, with a regression test. Found and fixed as part of the IndexTTS-2 installer work. (#1083)

CI

Test-suite debt sweep. Exports coverage (26 new tests, which caught two real router bugs), fp16 default-dtype leak instrumentation, and a suite-order pollution class root-caused at its source — the checks that guard every release got stricter. (#1081)

Windows x64 artifacts

f7d26a94af61fad908c21adb9da8f63196f8de58bdd4900562c09b0838c11348 *OmniVoice Studio_0.3.18_x64_en-US.msi
d05d99953b477ab59271b4671b91665bf40802bb9b33e27e29ff51cbf80c83a0 *OmniVoice Studio_0.3.18_x64_en-US.msi.sig

Linux x64 artifacts

f83cc7bd935ebabfba77872da0290fd91863d9ae74d65948bcb3d1742332a465  OmniVoice Studio_0.3.18_amd64.AppImage
aee0e0df4c01032b3bd0bd07a16867ac9556dad179a6095bc921f5ff5efee862  OmniVoice Studio_0.3.18_amd64.AppImage.sig

macOS Intel artifacts

0f6191a23650030feee045f8aa7290d21b7f3f4689e53f4cbc3ae0ff21c927cc  OmniVoice Studio_0.3.18_x64.dmg
1eef1bba602e593306e1da2d0b4a52e8e8aa5a96c3cce523489611d58aa8deb7  OmniVoice Studio.app.tar.gz
1961333bd38990ccc78ad110e74d6ba5e16161d725fe4e2bed12be3ecb4e4c43  OmniVoice Studio.app.tar.gz.sig

详细ChangeLog

Github榜单

日榜

2026-09-07

rank🥇

1205

周榜

2026年第37周

rank4

5104

欢迎回来

VoiceStudio 本地化语音生成工具

VoiceStudio 本地化语音生成工具

为VoiceStudio打分

收藏到合集

用户评论 (0)

#1274.[Bug] Not working on Strix Halo Radeon 8060s

#35.OmniVoice Studio Setup failed

#3.Native Desktop Applications

#116.[Bug] transcription prodused no segments. no module pkg resources

#4.installation problems

#961.[Install] Ubuntu 26.04 white screen at first launch

#255.[Bug] Transcribe stream dropped before emitting any segments

#1348.[Bug] TTS generate ran for more than 300s of actual compute time and was abandoned — t

#1112.[Bug]

#1000.[Bug] Can't reach the local OmniVoice backend — it may still be starting up, or it sto

#1748.iCON INT

#1704.Voice Changer With Audio File input

#1631.help with CosyVoice 3 en VoiceStudio, doesn´t work!

#1558.not available, even though it is already installed.

#1554.Deep-dive: three reviews, three races — making the crash supervisor safe against its own retry flow

#1553.Deep-dive: binding the socket in ~1 second — and the shutdown race hiding inside deferred startup

#1552.Deep-dive: why every Rust test binary died on Windows before running a single test (STATUS_ENTRYPOINT_NOT_FOUND)

#1286.PocketTTS (Kyutai) as a fast CPU-only engine?

#1091.Problem loading model

#593.Wrong Gender Voice

v0.5.1

Changed

Added

Docs

Fixed

CI

Linux x64 artifacts

macOS Intel artifacts

Windows x64 artifacts

Contributors

v0.5.0

Changed

Added

Docs

Fixed

CI

macOS Intel artifacts

Windows x64 artifacts

Linux x64 artifacts

Contributors

v0.4.2

Changed

Fixed

Linux x64 artifacts

macOS Intel artifacts

Windows x64 artifacts

Contributors

v0.4.1

Changed

Added

Docs

Fixed

macOS Apple Silicon artifacts

Windows x64 artifacts

macOS Intel artifacts

Contributors

v0.4.0

Changed

Removed

Added

CI

Docs

Fixed

macOS Apple Silicon artifacts

macOS Intel artifacts

Windows x64 artifacts

Contributors

v0.3.22

Added

Fixed

Windows x64 artifacts

Linux x64 artifacts

macOS Intel artifacts

v0.3.21