github debpalash/VoiceStudio v0.5.9-beta.6
v0.5.9-beta.6 — VoiceStudio

pre-releaseone hour ago

Download

Platform Download Signing
Windows x64 Installer ⚠️ Unsigned: SmartScreen shows "Windows protected your PC"; choose More info → Run anyway
Windows ARM64 (experimental) Installer ⚠️ Unsigned: SmartScreen shows "Windows protected your PC"; choose More info → Run anyway
macOS Apple Silicon DMG ✅ Signed with Developer ID and notarized by Apple
macOS Intel DMG ✅ Signed with Developer ID and notarized by Apple
Linux x64 AppImage · deb — Linux packages are not code-signed
Android APK · AAB ✅ Signed APK (signature verified); AAB is the Play upload bundle
iPhone Unsigned IPA ⚠️ Unsigned: sign it with your own Apple ID to sideload; not a TestFlight build

Highlights

  • GTX 900/1000-series NVIDIA cards run on the GPU again with a CUDA 12.6 PyTorch build (#4242)
  • Close to tray: keep VoiceStudio and its jobs running when you close the window (#4174)
  • Download buttons work in the Docker/web build (#4224)
  • Faster-feeling starts: no more "first run" hint on every launch, and slow cold loads are no longer killed after 300 s (#4228)
  • Security hardening across file actions, HTTP errors and CI, plus clearer Windows setup errors (#4173–#4188, #4213, #4230)

Changed

  • Opt-in automatic failure reports now name which part of the app failed (main process, GPU or helper process, GPU job timeout or network) instead of reporting an unknown error (#4243)
  • Removed the unused retired desktop release cleanup helper and its ownership entry. (#4172) — thanks @salemalkadadi-debug!
  • Removed an unused legacy update-busy helper while retaining active restart protection (#4149) — thanks @themrbeasley!
  • Retired an unused legacy update toast while preserving the active Electron update controls (#4141) — thanks @themrbeasley!
  • Retired the unreachable legacy shared UI and the guard tests that held it, repointing the live-behaviour guards to the renderer, and dropped 12 unused npm packages (#4095)
  • Smaller install: removed unused leftover images and old helper code from the app source (#4087)
  • Projects grid view stays smooth with thousands of items by drawing only the cards in view (#4079)
  • Long Projects, dub project and Transcribe history lists now render only the rows on screen, so they stay fast with thousands of items (#4074)
  • Idle CPU: a hidden or minimised window no longer runs polling timers or elapsed-time clocks, and they refresh once when it is shown again (#4073)
  • Removed unused renderer files and 477 unused translation strings, and locale:check now fails on an English key no code uses (#4047)
  • Removed more unused legacy shared components, utilities and API helpers from the old shared UI folder (#4045)
  • Local-model folder saves no longer unload the engine when the stored folder did not change (#4043)
  • Removed more unused legacy settings, gallery and engine screens from the old shared UI folder (#4042)
  • Idle polling is relaxed: the backend health probe slows when steady, cloud-sync polling pauses while the window is hidden, and the Windows parent watcher polls less often (#4041)
  • Removed more unused legacy dubbing screens and hooks from the old shared UI folder (#4040)
  • Release notes now open with one Download table listing every desktop and phone download with its signing, followed by the Highlights (#4037)
  • Removed about 25,000 lines of unused legacy interface code, two unused font packages and the always-skipped weekly evals job; the app itself is unchanged (#2603, #4032)

Added

  • Windows and Linux x64 PCs whose NVIDIA cards are all GTX 900/1000-series (compute capability 5.0 to 6.x, driver 525 or newer) now run on the GPU with a CUDA 12.6 PyTorch build, a one-time ~3 GB download on the next launch; voices, projects and models are kept (#4242)
  • Settings → General can keep VoiceStudio and its running jobs alive in the tray when you close the window, instead of quitting (#4174)
  • Stories can merge adjacent lines with the same speaker and compatible voice/speed settings while preserving text and inline directions — thanks cvak on Discord! (#4161)
  • Added separately opted-in failure reporting through Supermetry app intake across desktop, mobile and server installs, with private metadata and consent-controlled retries (#4137)
  • Dubbing exports the selected transcript as plain UTF-8 TXT, including edited text before generating a dub, with native Save As and paired phone sharing. (#4155) — thanks @shivamdhiman999111-source!
  • A dub interrupted by a crash or restart now picks up from its finished segments instead of starting over (#4088)
  • omnivoice-gguf now supports voice design, non-verbal tags and the decode-steps slider (#4084) — thanks @snazzybean!
  • Cebuano is now a dub translation target (#4058, #4072) — thanks @onparr-24!
  • Home shows a What's new link, winter pixel snow, a Christmas-week Santa, two rows of contributor portraits, and an easter egg that hides the credits until the next update after you visit three contributor profiles (#4070)
  • Voice Cloning has a generation queue: takes started while one is rendering wait their turn and run one after another — thanks @ItzSudhan! (#4064)
  • Audiobooks and Stories have a New project button that starts an empty draft without touching saved projects (#4055)

Docs

  • The localization contract no longer exempts the removed legacy shared i18n catalog, which pointed reviewers at a path that does not exist (#4186) — thanks @DeryFerd!
  • Document the enforced up-to-date merge gate while the repository cannot use GitHub merge queues (#4165)
  • Track the unresolved VoxCPM2 installation report separately from GPU fallback fixes, reported by vinhson (#4154)
  • Track the unresolved Windows model-loading crash and distinguish merged installer repairs from released fixes (#4146) — thanks @renaldyramadani123-code!
  • Installation tracking separates shipped fixes, pending repairs and unresolved report evidence (#4121)
  • Document host-managed Tailscale access for Docker without sharing the host daemon socket (#4111) — thanks @MaverickKB!

Fixed

  • Windows setup now explains a torch c10.dll "WinError 1114" load failure, naming an outdated Visual C++ runtime when that is the cause and listing what to try otherwise (#4230)
  • Trimmed speech now ends with a short fade-out instead of a sharp cut-off (#4229)
  • The startup screen only says "first run can take several minutes" while the Python environment is actually being installed, and the OmniVoice engine now reports progress during cold imports, reference transcription and synthesis so slow starts are no longer killed as "sent nothing for 300s" (#4228)
  • Windows ROCm setup prefers the AMD driver's newer HIP runtime, fixing an amdhip64_7.dll crash on dGPU + Ryzen iGPU systems, with automatic fallback to the bundled runtime (#4225) — thanks @giuseppefaro!
  • Download buttons for generated audio, batch and audiobook exports now work in the Docker/web build (#4222, #4224) — thanks @msy-jc!
  • Windows startup errors distinguish Smart App Control evaluation mode from an active policy block (#4213) — thanks @isnaenihidayat!
  • Safari extension errors with WebKit top-level stack frames no longer open the repair and bug-report prompt (#4211) — thanks @karlnorling!
  • Dubbing keeps API requests responsive while marking fresh speech on the watermark worker; recovered speech retains its existing mark (#4198) — thanks @swax!
  • Stories and Audiobook keep saved audio with its original connection and offer an explicit availability check for older results (#4196)
  • The changelog reference allowlist no longer outlives the entries it exempts: two dead exemptions were removed and a guard now reports any future ones (#4188) — thanks @DeryFerd!
  • HTTP error messages no longer interpolate raw exception text: filesystem and parsing failures return fixed, actionable text, and the remaining details pass through core.scrub.scrub_text to redact secrets and home paths (#4183) — thanks @DeryFerd!
  • Release preparation no longer aborts when the contributors block cannot be assembled, and a missing previous release only skips that block (#4181) — thanks @DeryFerd!
  • The commit identity check now resolves to a definite status when a pull request moves mid-run, instead of leaving a required check pending forever (#4180) — thanks @DeryFerd!
  • The gallery pin check now validates a pull request with the checker from its base commit, so a PR cannot move the pin and weaken the check in one commit (#4179) — thanks @DeryFerd!
  • Bundled installs that cannot run database migrations now enforce one cloud-sync identity per voice, so a duplicate identity is caught locally instead of surfacing later as a sync conflict (#4176) — thanks @DeryFerd!
  • Phone model catalogues generate on Windows again: the hash-pinned and byte-compared contract artifacts keep LF endings in every checkout (#4175) — thanks @DeryFerd!
  • Restrict native file-opening actions to the primary app window while preserving trusted recorder exports and window-local controls (#4173) — thanks @DeryFerd!
  • Community Gallery limits mounted cards in large catalogues while retaining the current preview, keyboard focus and paging (#4169)
  • Desktop failure observers retain the original consent during backend restarts and preserve normal Promise behavior. (#4209)
  • CosyVoice catalogue downloads skip unused alternative model files and use the same required assets as managed installs. (#4207)
  • Android keeps large manuscripts editable and recoverable, and retains reporting consent through permission prompts. (#4206) — thanks @gaino-svg!
  • The closed-PR CI sweep requests cancellations incrementally, preserves partial progress and reports incomplete searches or skipped requests accurately (#4189) — thanks @DeryFerd!
  • Kept automatic error reports bound to original consent through delayed work and recovery, and preserved native CI verification (#4184) — thanks @DeryFerd!
  • Windows ROCm setup preserves working-install stamps, verifies the pinned interpreter, and reports native crash exit codes (#4182) — thanks @DeryFerd!
  • Windows checkouts no longer fail the CLI packaging tests: the Debian maintainer-script cases skip when the host has no POSIX shell or symlink privileges (#4178) — thanks @DeryFerd!
  • Phone dubbing clip/stem exports check output space on both platforms and avoid false low-space errors on iOS (#4168)
  • Linux ROCm setup skips CUDA downloads and preserves its installed PyTorch stack on resume (#4167) — thanks @khryztiam!
  • GPU readiness and in-process routing reflect completed self-test failures (#4166) — thanks @khryztiam!
  • CLA audits keep exempt accounts visible for provenance review without reporting a missing required signature (#4164)
  • Release rehearsals show measured Android APK signing, and tagged downloads require matching signature evidence (#4163)
  • Manual Windows crash reports retain the faulting thread ahead of verbose status tails — thanks @renaldyramadani123-code and @freemood36-svg! (#4162)
  • macOS publishing checks Gatekeeper even when Windows installers are explicitly unsigned (#4153)
  • The reviewed model-install steps now require the same authenticated consumer as the rest of the model manager (#4152) — thanks @DeryFerd!
  • Diagnostic bundles and Electron bug reports redact share PINs and encoded credential query parameters (#4151) — thanks @DeryFerd!
  • Make the installed-model licence action open its terms and acknowledgement dialog — thanks hugotriunfa18 on Discord! (#4147)
  • Android model downloads resume through VPNs while preserving Wi-Fi policy and saved ranges (#4140) — thanks @JerryTimi!
  • Stopped playback errors from incorrectly claiming that generated audio files are missing (#4138) — thanks @AAYUSH-CODER1904!
  • Transcriptions can save clean plain text and timed SRT captions, with an explanation when subtitle timings are unavailable (#4135)
  • Cloud sign-in now rejects a discovery document whose token or revocation endpoint is not on the issuer's own origin (#4133) — thanks @DeryFerd!
  • Segmented model downloads pin public destinations with bounded DNS checks across redirects, ranges and proxies, preventing private-network fallback and CDN credential leaks (#4132) — thanks @DeryFerd!
  • The LM Studio model probe no longer sends the stored API key in clear text to a remote host unless plaintext LAN is allowed (#4131) — thanks @DeryFerd!
  • A deterministic test now fails when a new mutating API route ships without an authorization decision (#4130) — thanks @DeryFerd!
  • Docker deployments now require the administrator API key for the runtime log, log stream, recent errors and crash report, as the install docs already said (#4129) — thanks @DeryFerd!
  • MCP tools preserve remote caller permissions across local and HTTP backends, including failed transport lookups (#4128) — thanks @DeryFerd!
  • TTS and event WebSockets now enforce the same loopback/credential rule as dictation, closing a share-PIN bypass (#4126) — thanks @DeryFerd!
  • Imported project bundles can no longer make ffmpeg fetch remote URLs through a playlist-shaped media file (#4125) — thanks @DeryFerd!
  • Deleting history, rewriting dub scripts and clearing model-licence receipts now require an administrator, as the API docs already promised (#4124) — thanks @DeryFerd!
  • SoniTranslate installation and start/stop now require a local administrator, like every other engine installer (#4123) — thanks @DeryFerd!
  • Model deletion over the API now requires administrator rights (#4122) — thanks @DeryFerd!
  • Crash recurrence reports recognize current Electron version fields and distinguish beta releases (#4120)
  • Reclaim idle GPU models before generation when VRAM or RAM is low (#4117) — thanks @vandyahmad24!
  • Model downloads reject damaged files and retry without reusing corrupt ranges, while cancellation preserves verified work (#4116)
  • Reuse verified setup wheels on low-space retries and handle interrupted size probes without an uncaught error. (#4115)
  • Saved and inline pronunciation choices apply consistently across speech, dubbing, batch and voice-conversion routes (#4114) — thanks @nntloc!
  • Stop cancelled setup retries promptly and keep verbose runtime probes from blocking on their diagnostic output. (#4113)
  • Long recordings use duration-scaled transcription limits across uploads, dubbing, child engines and MCP tools (#4112) — thanks @xabiherdz-svg!
  • URL imports restore their connection guard after temporary DNS resolver changes (#4110)
  • Update the optional Dagger pipeline's OpenTelemetry dependencies to fix excessive CPU use when parsing malformed trace context. (#4109)
  • Optional analytics no longer delays backend readiness, and consent changes invalidate queued startup work (#4108) — thanks @cf3901646!
  • Auto setup can retry PyPI failures through the regional mirror and report the actual failed host (#4107) — thanks @mrkeshi!
  • Malformed model-install status responses keep the interface open and preserve the last valid download state (#4106) — thanks @ftlcustomerservice73-ui!
  • Windows installer process checks skip custom PowerShell profiles, and source installs use shorter build paths to avoid MSVC linker failures (#4099)
  • Restarting to install an update keeps update status available while the local engine shuts down, avoiding missing-handler errors (#4098)
  • Automatic blank-renderer recovery keeps the open workspace instead of returning Stories or Audiobook to Home (#4097)
  • Dubbing keeps the saved recording for cloned and designed voices with style instructions, so each segment uses the selected speaker's reference (#4093)
  • Setup on Auto-detect now retries a stalled Python download (cpython-… (download)) once from the other source, GitHub or its mirror, instead of stalling again (#4085)
  • A refused cross-site API request now names its origin and gives the exact OMNIVOICE_ALLOWED_ORIGINS line to add, so other local apps such as AIRI can be connected (#4083)
  • Setup no longer crashes with "Cannot set property code" when a runtime download or check is cancelled on Windows (#4082)
  • MOSS-TTS-Nano speaks again on a fresh one-click install: the transformers 5 shim no longer breaks model loads in its pinned transformers 4 environment (#4063, #4078) — thanks @RerankerGuo, and @rockziepromax for the report!
  • OmniVoice GGUF falls back to the CPU when the GPU backend crashes, for example Vulkan on a GTX 1080 (#4050, #4072) — thanks @MeinLive!
  • Full dubs no longer fail when translation fills segments with free-text directions like "slow"; unsupported ones are dropped for that segment (#4054, #4072) — thanks @user555gtb!
  • Audiobook MP3 exports play and download again instead of returning 404 (#4071, #4072) — thanks @dimfk001git!
  • Colab: setup cells show the failing command's output, the security-floor check judges the Mako copy Python imports, and the web UI loads through the new *.prod.r.colab.dev proxy — thanks @hkarmoush! (#4069)
  • Turning on LAN sharing no longer crashes the backend with an Internal Server Error — thanks @blutuxyz! (#4059)
  • The OmniVoice GGUF engine finds its bundled runtime again in installed apps instead of reporting it missing — thanks @b84330808! (#4053)
  • When the operating system times out starting the engine (spawn ETIMEDOUT, seen on macOS), VoiceStudio retries twice and then says to restart the computer (#4039)
  • Settings → Updates lists only your channel's releases, and Beta suggests switching to Stable when a newer stable is out (#4038)
  • SSML [spell] keeps flag, ZWJ-joined and skin-tone emoji whole instead of splitting them (#4020) — thanks @kevin9327!

CI

  • CI actions gain artifact and no-push Docker contract checks, with one bounded mobile artifact-upload retry (#4192)
  • CI pins the version of the security analyser it installs at run time, so a new release cannot silently change the findings (#4187) — thanks @DeryFerd!
  • Documentation and inventory changes run a read-only docs drift check on relevant pull requests, with rolling-issue maintenance kept in nightly/manual jobs (#4185) — thanks @DeryFerd!
  • Container builds grant the registry-write token only to the jobs that publish images, not to every job in the workflow (#4177) — thanks @DeryFerd!
  • CI reuses unchanged iOS checks, rechecks shared diagnostic inputs and skips Nix builds for documentation and test changes. (#4208)
  • Windows release and optional-engine tests use portable text fixtures, native Bash and normalized cache paths (#4197) — thanks @swax!
  • Release builds load app-intake credentials from GitHub Actions Secrets without public-variable fallbacks (#4204, #4205)
  • Fixed Android screenshot capture waiting behind the optional error-report prompt (#4202) — thanks @velixio!
  • Keep iOS voice-search and download checks reliable across saved model choices (#4171)
  • The iPhone smoke test waits for typed text to reach the editor instead of reading it once, removing a CI flake (#4100)
  • MOSS compatibility tests keep checking the active Transformers version after another test restores imported modules (#4096)
  • The iPhone UI smoke test keeps its slow-inventory case loading for the whole test, so a slow simulator no longer fails it (#4076)
  • Docker's web build no longer fails the unused-key check when only electron/ is copied, and the iPhone CI job allows for a native-runtime cache miss (#4052)
  • Removed unused scripts, the old Tauri MSI tooling and leftover Turborepo ignore lines (#4046)
  • Accessibility smoke timeouts now name the requests still pending, so a flaky CI failure points at its cause (#4036)

Bug reporters

Contributors

New since v0.5.9-beta.5: @debpalash, @DeryFerd, @swax, @isnaenihidayat, @karlnorling, @x3vyx, @salemalkadadi-debug, @velixio

Everyone who contributed since v0.5.8: @debpalash, @DeryFerd, @velixio, @ItzSudhan, @swax, @b84330808, @blutuxyz, @isnaenihidayat, @karlnorling, @kevin9327, @RerankerGuo, @x3vyx, @hkarmoush, @salemalkadadi-debug, @vandyahmad24

Don't miss a new VoiceStudio release

NewReleases is sending notifications on new releases.