* fix(offline): patch transformers mistral-regex check to survive HF failures
transformers 4.57.x's `PreTrainedTokenizerBase._patch_mistral_regex` calls
`huggingface_hub.model_info(repo_id)` unconditionally during any non-local
tokenizer load to probe for Mistral-family models. The call raises on
`HF_HUB_OFFLINE=1`, on network outages, and on slow/blocked HF endpoints,
and transformers doesn't catch any of it — the exception bubbles out of
`from_pretrained` and kills the load for unrelated engines (Qwen TTS,
Qwen CustomVoice, TADA, etc.).
0.4.2's load-time `force_offline_if_cached` guard walked straight into
this trap: on cached online users it flipped `HF_HUB_OFFLINE=1` and
converted a healthy load into a hard crash. 0.4.3's inference-path guard
masked it; #524 removed the inference guard in 0.4.4, and users updating
to 0.4.4 started hitting the same error on the load path instead
(#526).
Fix:
- Wrap `_patch_mistral_regex` so any exception from the inner HF
metadata check is swallowed and the tokenizer is returned unchanged.
Voicebox never loads Mistral models, so the regex rewrite this check
gates is a no-op for us; matches the success-path behavior for
non-Mistral repos (tokenization_utils_base.py:2503).
- Drop the `force_offline_if_cached` wraps from every load path
(pytorch_backend Qwen + Whisper, qwen_custom_voice_backend,
mlx_backend Qwen + Whisper). With the mistral patch in place they
provide zero value and only risk re-introducing the same class of
bug. Helper and its unit tests stay — still correct for targeted
future use.
- Add `backend/tests/test_offline_patch.py` covering
OfflineModeIsEnabled / ConnectionError suppression, success
pass-through, idempotence, and the missing-method no-op path.
Fixes#526.
* fix(offline): install mistral-regex patch for non-MLX backends
The previous commit left the patch wired only through ``mlx_backend.py``'s
existing import of ``hf_offline_patch``. On Windows/Linux/CUDA users who
never load the MLX backend (everyone who hit #526), the patch module was
never imported, so ``patch_transformers_mistral_regex`` never ran and the
crash persisted.
Hoist the import into ``backends/__init__.py``. Every backend imports from
this package, so the module-level patch install runs before any
``from_pretrained`` call regardless of which engine the user picks.
Caught by CodeRabbit and Cursor Bugbot on #530.
0.4.3 wrapped every inference body (`generate`, `transcribe`,
`create_voice_clone_prompt`) with `force_offline_if_cached(True, …)` to
prevent lazy HF lookups from hanging when the network drops
mid-inference (#462). That trade broke online users: the guard flips
`huggingface_hub.constants.HF_HUB_OFFLINE` globally, so any legitimate
metadata call the library makes during generation (e.g. revision
resolution via `HfApi().model_info`) now raises:
Cannot reach https://huggingface.co/api/models/Qwen/Qwen3-TTS-…:
offline mode is enabled.
Hit by multiple users on 0.4.3 within hours of release. The offline
blast radius is much larger than the original hang it fixed.
This reverts the inference-path guards. Load-path guards stay — those
worked fine in 0.4.2 and aren't the source of the regression. The
`force_offline_if_cached` helper itself is unchanged; tests still pass.
The #462 hang (network dropping mid-inference) remains unaddressed by
this commit and will need a targeted fix that doesn't flip a global
flag — most likely per-call timeouts or library-specific
`local_files_only` arguments, not a process-wide env mutation.
* fix(release): notarize and staple macOS DMGs
Tauri's bundler signs the .app and notarizes it, but ships the .dmg
wrapper unnotarized. Gatekeeper rejects that on macOS 15 Sequoia
(caught by Homebrew Cask CI) and causes the 'app isn't signed'
dialog on older Intel Macs when Apple's notarization servers are
slow (issue #509).
New step submits each built DMG to notarytool, staples the ticket,
verifies with spctl, then overwrites the release asset tauri-action
already uploaded to the draft release.
Adds ~5-10 min per macOS job (notarytool round-trip).
* fix(release): fail loudly when no DMG is found to notarize
Empty glob + nullglob was silently skipping the loop body, so if
Tauri's bundler output path changed we'd re-publish the unnotarized
DMGs with a green CI. Assert the glob matched at least one file.
* fix(release): resolve release tag from tauri.conf.json, not GITHUB_REF_NAME
GITHUB_REF_NAME is the branch name when the workflow runs via
workflow_dispatch, so the gh release upload targeted the wrong thing
on manual runs. tauri-action derives its tag from tauri.conf.json's
version field via the v__VERSION__ template; use the same source so
the two always agree.
fugashi (pulled in by misaki[ja]) needs a MeCab dictionary at runtime.
The `unidic` package that ships today contains no data — it relies on
`python -m unidic download` (~526MB), which isn't run by `just setup`
and won't survive PyInstaller freezing.
Switch to `unidic-lite`, which bundles a MeCab-compatible dict inside
the wheel (~50MB). Collect its data files in build_binary.py so frozen
builds also pick up the dicdir. Same failure mode and same fix shape as
the existing en_core_web_sm pre-install.
* feat(i18n): add i18next foundation with English + zh-CN locales
Installs i18next + react-i18next + language detector and wires up a
language selector in the General settings page. Extracts strings from
the highest-visibility surfaces: all settings tabs, model management,
sidebar nav, main editor, and the floating generate box. Remaining
strings (profile forms, history, stories/effects/voices/audio tabs)
can land in follow-up PRs.
Closes#411.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(i18n): ensure language switch actually re-renders the tree
- `nonExplicitSupportedLngs: true` was normalizing `zh-CN` → `zh` in
some code paths; since we have explicit `zh-CN` resources, swap it
for `load: 'currentOnly'` which keeps the code as-is.
- `react: { useSuspense: false }` — react-i18next v17 defaults Suspense
on, which can silently suspend components mid-switch and look like
"nothing happens" to the user.
- Use `i18n.language` (the raw current code) instead of
`resolvedLanguage` in the selector so the dropdown always mirrors
what we just set.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize ProfileCard, HistoryTable, and relative dates
- ProfileCard: "No description", "designed" badge, aria-labels, and
delete dialog.
- HistoryTable: delete / clear-failed / import / effects dialogs.
- formatDate: switch date-fns `formatDistance` locale based on
`i18n.language` so "5 minutes ago" becomes "5 分钟前" under zh-CN.
HistoryTable now subscribes via useTranslation so the table
re-renders when language flips.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize Stories tab (list, content, dialogs, toasts)
Covers the title + "New Story" button, empty states, story row
metadata (item count, updated time), the create/edit/delete dialogs,
and all toast notifications. Also handles StoryContent: "Select a
story" placeholder, search popover, "Export Audio" button, and the
"Generating N audios" pending indicator.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize history item and story item dropdown menus
Covers the "..." action menu on both the History table (Play, Export
Audio, Export Package, Apply Effects, Regenerate, Delete) and on
individual story chat items (Play from here, Remove from Story).
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize Effects tab (list, detail, dialogs, toasts)
Covers EffectsList (title, "New Preset", section headers, preset
cards) and EffectsDetail (header buttons for Save / Save as Custom /
Delete, name/description fields, preview section, Save as Custom
dialog, all toasts).
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize EffectsChainEditor and built-in preset names
- Effect type labels (Chorus/Flanger, Reverb, Delay, Compressor, Gain,
High-Pass, Low-Pass, Pitch Shift) and every param label (LFO speed,
Modulation depth, Threshold, Ratio, etc.) go through
`effects.types.<type>.{label,params.<param>}` with the backend string
as defaultValue fallback.
- Chain-level controls: "Load preset…", "Add effect…", "Clear",
Power/Remove button titles.
- Built-in preset names + descriptions (Robotic, Radio, Echo Chamber,
Deep Voice) are translated client-side; user-created presets keep
their original names.
Backend keeps returning English — frontend intercepts and translates
via key lookup, defaulting to the backend string so unknown
effects/params don't break.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize Create/Edit Voice modal and audio sample panels
ProfileForm now routes its title, description, voice-source toggle
(Clone from audio / Built-in voice), field labels (Name, Description,
Language, Engine, Voice, Reference Text, Default Engine, Default
Effects), sample tabs (Upload / Record / System Audio), action
buttons, and every toast + Zod validation message through i18n.
Also covers the three AudioSample panels (Upload/Record/System) — the
choose-file / start-recording / start-capture call-to-actions, the
"N remaining" countdown, "Recording complete" / "Capture complete"
states, and the Play / Transcribe / Remove / Record Again buttons.
SampleList too — the "No samples yet" empty state, per-sample edit
mode, mini-player aria labels, Delete Sample dialog, and toasts.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize Audio Channels tab (list, dialogs, device picker)
Covers the "Audio Channels" title and "New Channel" button, the empty
state, per-channel section labels (Output Devices / Assigned Voices),
the Available Devices right pane with its three contextual hints, the
"No voices assigned" fallback, and both Create/Edit dialogs (titles,
descriptions, field labels, Select placeholders, and the "(default)"
badge).
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize Voices tab (table header, search, inspector)
VoicesTab now translates the "Voices" title, search placeholder,
"New Voice" button, all six table column headers (Name, Language,
Generations, Samples, Effects, Channels), the avatar alt text, and
the per-row channel MultiSelect (placeholder + "(Default)" suffix).
VoiceInspector routes its form labels through the existing
`profileForm.fields.*` keys, has its own "Default Effects" hint and
avatar/save toasts, and reuses the ProfileForm Zod validation.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): localize ProfileList unsupported-model note and empty state
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): add Traditional Chinese (zh-TW) locale
Adds a zh-TW translation with Taiwan vocabulary conventions
(e.g. 預設 / 儲存 / 載入 / 匯入 / 匯出 / 設定 / 檔案 / 伺服器 / 裝置 / 網路).
Registers it alongside en and zh-CN; the language dropdown picks it up
automatically from SUPPORTED_LANGUAGES.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(i18n): add Japanese (ja) locale
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(i18n): localize relative dates for ja and zh-TW
formatDate only mapped zh-CN, so history timestamps stayed in English for
ja and zh-TW users even after the rest of the UI translated. Extend the
switch to ja and zhTW from date-fns/locale.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(i18n): address PR review feedback
Bugs:
- GeneralPage: network access toast used the keep-server-running title key
(wrong semantic scope). Add networkAccess.updatedTitle and use it.
- GeneralPage: fallback "Unknown" version was stored as a translated string
in state, so it stayed stale across language switches. Store null, resolve
the label at render time.
- GeneralPage: memoize the zod resolver on t and retrigger validation when
the locale changes so existing error messages retranslate.
- GpuPage: adding t to the CUDA progress EventSource effect deps caused the
SSE connection to be torn down and reopened on every language change,
potentially dropping in-flight download events. Capture t in a ref.
- HistoryTable: Effects dialog still rendered English "Source" / "Select
source version" / "Cancel" / "Apply" / "Applying..." — localize them.
- Locales: zh-CN / zh-TW / ja devSuffix was missing the leading space before
"(开发版)"/"(開發版)"/"(開発版)", so dev builds rendered "v0.4.2(开发版)"
instead of "v0.4.2 (开发版)".
Nits:
- ModelManagement: rename .find((t) => ...) callback param to avoid
shadowing useTranslation().t.
- GenerationPage: rename chunkLimit.value interpolation key from count →
chars so i18next doesn't silently activate pluralization if a translator
later adds _one/_other forms.
- LanguageSelect: narrow onValueChange handler param to LanguageCode.
Key count now 559 across en/zh-CN/zh-TW/ja (added 4 effectsDialog keys
plus networkAccess.updatedTitle).
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
mlx-audio's STT path imports miniaudio, but we install mlx-audio
--no-deps to dodge its transformers>=5.x pin. Nothing else pulls
miniaudio transitively, so fresh Apple Silicon installs fail to
transcribe with ModuleNotFoundError: miniaudio. Listed explicitly
and updated the stale comments in requirements-mlx.txt and
release.yml that claimed it came from other engines.
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
* fix(offline): guard inference paths with HF_HUB_OFFLINE (#462)
PR #443 wrapped the model *load* path with `force_offline_if_cached` so
cached models don't phone home at startup. The context manager restores
`HF_HUB_OFFLINE` on exit, which left inference paths (generate,
transcribe, voice-prompt creation) unguarded — and `qwen_tts`,
`mlx_audio`, and `transformers` perform lazy tokenizer/processor/config
lookups during inference. With internet on, those lookups are
near-instant and invisible; with internet off, `requests` hangs on DNS
or connect until the network returns. This is exactly what users in
#462 describe: model shows "Loaded", internet drops, generation
"thinks" forever, internet comes back, generation completes.
Chatterbox and LuxTTS don't exhibit this because their engine libs
resolve everything through already-cached paths at load time.
Fix: wrap each inference-sync body with `force_offline_if_cached(True,
...)`. Since inference only runs after a successful load, weights are
known to be on disk, so `is_cached=True` is unconditional.
Also adds the load-time guard that was missing from
`qwen_custom_voice_backend.py` — CustomVoice previously had no offline
protection at all.
Paths patched:
- PyTorchTTSBackend.create_voice_prompt (create_voice_clone_prompt)
- PyTorchTTSBackend.generate (generate_voice_clone)
- PyTorchSTTBackend.transcribe (Whisper generate + decoder-prompt-ids)
- MLXTTSBackend.generate (mlx_audio generate, all branches)
- MLXSTTBackend.transcribe (mlx_audio whisper generate)
- QwenCustomVoiceBackend._load_model_sync + generate
Does not address the secondary `check_model_inputs() missing 'func'`
error reported in the same issue — that's a `transformers` 5.x
version-skew bug on the install path, separate concern.
Fixes#462.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(offline): mutate cached HF constants + threadsafe refcount
Review feedback on the initial fix surfaced two real issues:
1. ``os.environ`` toggles alone don't flip offline mode.
``huggingface_hub.constants.HF_HUB_OFFLINE`` is read once at import
time into a module-level bool; ``transformers.utils.hub._is_offline_mode``
mirrors that bool at its own import time. The hot paths
(``_http._default_backend_factory`` in huggingface_hub,
``is_offline_mode`` in transformers) read the cached bools — not the
env — so mutating only ``os.environ`` was a no-op.
2. Race condition on concurrent inference. Two threads running inside
``force_offline_if_cached`` via ``asyncio.to_thread`` could have
thread A's ``finally`` strip thread B's offline protection mid-run.
Rewrite the helper to:
- mutate ``huggingface_hub.constants.HF_HUB_OFFLINE`` and
``transformers.utils.hub._is_offline_mode`` directly
- refcount concurrent users under a single ``threading.RLock`` so a
shared offline window is restored only when the last caller exits
- still write ``os.environ`` for anything that reads it dynamically
Also addresses the unused-variable ruff flag on the Whisper transcribe
path (``audio, sr`` → ``audio, _sr``).
New unit tests cover the cached-constant mutation, env propagation,
no-op on ``is_cached=False``, nested contexts, and a threaded race
where a slow thread must retain offline mode after a peer exits.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(offline): atomic entry rollback + tidy test assertions
Review follow-up:
- Wrap the `_offline_refcount == 0` setup in a try/except so any failure
during the cached-constant mutation (including unexpected non-ImportError
like RuntimeError or AttributeError from a half-initialized module)
rolls back *all* partial state before re-raising. Without this, a
mid-setup crash could leave `huggingface_hub.constants.HF_HUB_OFFLINE`
mutated but the refcount at 0 — a persistent offline flag outliving
the process.
- Swap ruff-flagged Yoda comparisons in the new test file (SIM300) and
add a module-level note warning that these tests mutate global state
and are not safe under cross-process parallelism.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* test(offline): make concurrency test deterministic and bounded
Replace the `sleep(0.15)` ordering hack with an explicit `threading.Event`
the fast thread sets in `finally`. The slow thread waits on that event
(bounded), then observes the flag — so we deterministically verify the
slow thread still sees offline mode after the fast thread has exited.
Also add timeouts to `barrier.wait()` and assert `not thread.is_alive()`
after the joins so the test can't hang on an unexpected failure path.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
* feat(ci): re-enable Linux release builds (CPU, deb+rpm)
Linux shipped briefly in March 2026 (b580189..103e98b) then was removed
with the message "github runners suck." The post-mortem: standard
ubuntu-22.04 hit disk-pressure during pip+PyInstaller, and a namespace
custom-runner attempt proved flaky. Nothing in the app itself was the
problem — the NVIDIA-package exclusions in build_binary.py (~3 GB
shaved from CPU builds), the CPU-only torch install order, the
PulseAudio/PipeWire audio capture code, and the Tauri deb/rpm bundle
targets all still work.
This restores ubuntu-22.04 to the release matrix as a CPU-only Linux
build with a disk-space cleanup step (jlumbroso/free-disk-space) to
address the root cause of the March failures. Ships .deb + .rpm only
— AppImage was explicitly dropped in e18757b due to glibc portability
issues, keeping that decision.
CUDA-for-Linux is intentionally deferred to a follow-up PR: the
GpuAcceleration.tsx frontend hard-codes CUDA download as Tauri-only
without a Linux branch, and AMD users already get ROCm acceleration
for free on a stock CPU-torch install (backend/app.py:148-160). The
NVIDIA-on-Linux case is the only remaining gap and is non-blocking
for a v1 Linux ship.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* chore(release): bump version 0.4.1 → 0.4.2
tauri-action uses tauri.conf.json's version to name the release, so
pushing v0.4.2 as a tag alone was insufficient — the workflow was still
trying to publish to v0.4.1 (immutable) and failing. Bumps all workspace
package.json files, Cargo.toml, Cargo.lock, and tauri.conf.json.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* ci(release): add Linux bundle-step watchdog + verbose cargo logging
The first v0.4.2 attempt hung inside tauri-action at the bundling stage
on ubuntu-22.04 for ~28 min with no output. Same failure mode that drove
the March 2026 removal (commit 103e98b). Without visibility we're
guessing — probable causes include linuxdeploy/AppImage download stalls
(despite --bundles deb,rpm), cargo link on a cold cache, or disk
pressure during the final link step.
- timeout-minutes: 30 on tauri-action for Ubuntu (45 min elsewhere) —
fail fast with logs instead of waiting out the 6hr job timeout.
- --verbose added to the Linux build args so cargo streams progress.
- CARGO_TERM_VERBOSE + RUST_BACKTRACE=1 exported on tauri-action.
- New 'Disk / environment snapshot' step dumps df/free/tool versions
right before the tauri step so we can correlate with any later
OOM/ENOSPC failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* ci(release): disable Tauri updater artifacts on Linux (stops linuxdeploy hang)
Round-2 diagnosis from the v0.4.2 attempt: cargo finished in 3m 55s,
.deb and .rpm were bundled within 22s, then the step hung silently for
25 min until our 30-min watchdog killed it — no further log lines.
tauri.conf.json has `createUpdaterArtifacts: "v1Compatible"`. On Linux
the v1-compatible updater path wants a .AppImage.tar.gz, which means
downloading linuxdeploy-x86_64.AppImage from GitHub at build time.
That download is the silent blocker — same signature as the March 2026
"github runners suck" removal (commit 103e98b).
Fix: pass `--config {"bundle":{"createUpdaterArtifacts":false}}` on the
Linux build only. Mac/Windows continue to produce signed updater
artifacts as before. Linux users update via apt/dnf; Tauri in-app
auto-update for Linux can come later (and would require shipping
AppImage alongside deb/rpm).
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* ci(release): pin free-disk-space action + stop removing LLVM
Review feedback on the Linux release workflow:
1. `jlumbroso/free-disk-space@main` was an unpinned ref in a job that
runs with `contents: write` and handles signing secrets. Pin to the
v1.3.1 commit SHA so a force-push or repo compromise can't inject
arbitrary code into our release flow.
2. `large-packages: true` runs `apt-get remove '^llvm-.*'`, wiping LLVM
just before the next step installs `llvm-dev`. That wastes CI time
and risks cascade-removal of reverse deps that won't be pulled back
in by `llvm-dev` alone. The remaining toggles (android, dotnet,
haskell, swap-storage) already clear ~20 GB, which is enough
headroom for the Python + torch + PyInstaller build.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
* fix(audio): preprocess reference samples instead of rejecting them
Uploaded/recorded voice samples were rejected outright whenever the peak
exceeded 0.99 ("Audio is clipping (reduce input gain)"). That wasn't
actionable: a recording in the app has no pre-gain control, and an
already-captured file can't be re-taken by the user. The Settings
"Normalize audio" toggle only affects generated TTS output, so users who
enabled it expecting it to help with sample uploads were still blocked.
Replace the hard reject with a small, always-on preprocess step that
runs right after load:
- DC-offset removal
- Conservative edge-silence trim (top_db=30) with 100 ms padding kept
- Peak cap at 0.95 if the input peak exceeds that
Duration and RMS checks now run on the preprocessed waveform, so
samples that were previously rejected for being "hot" are accepted and
stored with safe headroom. True in-waveform clipping artifacts still
can't be repaired — peak scaling only prevents downstream re-clipping
during multi-sample combination and TTS inference.
Adds a unit-test file (previously none existed for audio.py).
Fixes#456.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(audio): raise trim threshold, cap pad at net-neutral
Review feedback on the preprocessor:
1. ``trim_top_db=30`` was labelled "conservative" in the docstring but is
actually *more* aggressive than librosa's default of 60. Normal
speech dynamic range sits around 30 dB, so 30 dB would eat quiet
trailing syllables and soft consonants. Raise the default to 40 dB —
below normal speech dynamic range but still catching obvious edge
silence — and fix the docstring.
2. Unconditional 100 ms edge padding ran even when ``librosa.effects.trim``
removed nothing. For a well-recorded 29.9 s upload that path would
push the waveform past the 30 s ceiling and trigger a spurious "too
long" rejection. Only pad when trimming actually shortened the
audio, and cap the pad so the output never exceeds the input length.
Adds a regression test for the net-neutral length behaviour.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
The 0.6B slot was aliased to the 1.7B repo as a temporary fallback
because `mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16` wasn't published
when MLX support shipped. That conversion is live now, so use it —
Apple Silicon users picking 0.6B get the actual 0.6B model (1.2 GB
instead of 3.5 GB).
Also drops the now-obsolete troubleshooting entry and updates the
triage notes in PROJECT_STATUS.md.
Fixes#485.
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
* fix(landing): use qwen_custom_voice in API example (instruct is CustomVoice-only)
The curl snippet showed engine: "qwen" alongside an instruct field, but base
Qwen3-TTS has no instruct path — that's a Qwen CustomVoice feature.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(landing): use a realistic UUID for profile_id in API example
Profile IDs are str(uuid.uuid4()), not slugs (see backend/services/profiles.py:175).
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* feat(landing): add polished /download page — no more dumping users on GitHub
Users were clicking download, landing on the GitHub releases page, and filing
confused comments along the lines of "I ended up on some blog site called
GitHub." We now route every download CTA through a dedicated /download page
that auto-triggers the platform-specific download and gives users a polished
post-click experience with donate + docs + AI help prompts.
- New /download page:
- Big app logo + "Your download has started" messaging.
- Auto-detects platform from ?platform=X or navigator.userAgent.
- Programmatically clicks a hidden anchor to trigger the file download
without leaving the page.
- Platform-specific buttons as a visible fallback for "download not
working" / manual-pick.
- Personal donate spiel + Buy Me a Coffee button.
- Resources grid: docs, DeepWiki ("got questions? ask AI"), GitHub.
- Landing page download section cards now link to /download?platform=X
instead of the asset URL directly.
- /download/[platform] (used by README/docs links) now redirects to the
/download page rather than straight to the asset or to GitHub on error.
- Drops unused downloadLinks state from the landing page.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(landing): use official platform brand icons via simple-icons
The hand-rolled Linux SVG path wasn't actually Tux — it was a symmetric
placeholder shape. Apple/Windows were close but not canonical either.
- Apple + Linux: pulled from @icons-pack/react-simple-icons (SiApple, SiLinux).
- Windows: simple-icons drops the Microsoft mark over trademark policy, so
the Windows 11 flag is inlined from Microsoft's public brand guidance.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(landing): route Download CTAs to /download page, not the section anchor
Hero CTA, navbar link, and footer link were all scrolling to #download
(the section at the bottom of the page) instead of going to the new
/download page that triggers the actual download.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* chore(landing): run dev server on Node instead of Bun runtime
Bun runtime + Next 16 Turbopack dev server intermittently trips a
JavaScriptCore allocator panic ('pas panic: deallocation did fail ...
Alloc bit not set') after a few requests. Dropping --bun keeps Bun as
the package manager but runs next dev on Node, which is stable.
Build + start keep --bun since one-shot invocations don't exhibit the
allocator drift.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(landing): route Linux users to /linux-install instead of attempting download
No prebuilt Linux binary exists yet (see /linux-install for build-from-source
instructions). The /download page previously treated Linux like the other
platforms — auto-triggering a non-existent AppImage and offering a dead
manual button.
- /download page: if platform resolves to 'linux' via ?platform or UA detect,
window.location.replace('/linux-install') — never try to auto-download.
- Manual Linux card: label changed to "Build from source" and links to
/linux-install (no download attribute, no asset URL).
- /download/linux pretty URL: 307s straight to /linux-install.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* docs: consolidate troubleshooting into the MDX docs site + status updates
- Delete docs/TROUBLESHOOTING.md; the canonical troubleshooting guide now
lives under docs/content/docs/overview/troubleshooting.mdx so it's served
from docs.voicebox.sh alongside the rest of the docs.
- CONTRIBUTING.md + README.md: repoint "Troubleshooting" references to the
new MDX path. README gets a top-level callout so users hit the guide
before filing an issue.
- PROJECT_STATUS.md: refresh issue/PR counts, document the flash-attn
warning (cosmetic on all platforms; CUDA-only, fallback is PyTorch SDPA
which is near-FA2 on Ampere+) with per-platform context + community
Windows wheels + SageAttention/xformers alternatives, add WebAudio
audio-session bug note (tracked separately in PR #486), and expand the
Qwen 0.6B→1.7B MLX fallback explanation for triage.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(landing): address PR #487 review feedback
- Preserve canonical camelCase platform aliases (macArm, macIntel) in the
/download/[platform] redirect so those URLs don't lose their platform param.
- Add accessible title + role="img" to the inline Windows SVG so it passes
Biome's a11y rule and announces to screen readers.
- On /api/releases fetch failure, show an explicit error state with a single
intentional link to GitHub releases — no more silent GitHub fallback or
disabled-button UX lie. Keeps normies off GitHub unless they opt in.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
Keep a silent looping <audio> element mounted at the app root so macOS
never tears down the CoreAudio session. Without this, backgrounding the
app long enough leaves WaveSurfer's AudioContext in a state where play()
resolves and timeupdate fires, but no audio reaches the output — and not
even cmd+R (full JS reload) restores it, only a full app relaunch.
Uses a zero-PCM WAV blob at full volume rather than a muted element,
since WebKit can optimize muted media away and defeat the purpose.
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
- Add three tutorial cards (Danish Sofi, StinkyScrublet, mikbes)
- Navbar: switch parent to flex/justify-between on mobile (grid on sm+),
unhide Donate button so both CTAs sit on the right, matching desktop
- Hero CTAs: keep Download and GitHub side-by-side on mobile instead of
stacking vertically
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
The previous fix (#481) capped transformers at 4.57.6 in requirements-mlx.txt,
but pip's clean resolver in CI can't satisfy that alongside mlx-audio>=0.3.1
(declares `transformers==5.0.0rc3` or `>=5.0.0`) — it backtracks through every
transformers and tokenizers version and exits with `ResolutionImpossible`.
The dev install worked only because mlx-audio 0.4.1 was already present, so
pip never tried to re-resolve.
mlx-audio 0.4.1 + mlx-lm 0.31.1 both declare transformers>=5.x but the API
surface we actually use works fine on 4.57.x in practice (verified across all
engines in dev). Install both --no-deps to bypass the resolver; transitive
runtime deps (huggingface_hub, librosa, numpy, numba, pyloudnorm, etc.) are
already pulled in by requirements.txt.
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
mlx-audio depends on `transformers` with no upper bound. Installing
requirements-mlx.txt after requirements.txt lets pip upgrade transformers
past the 4.57.x cap to 5.x, which breaks three engines in the frozen MLX
bundle:
- qwen-custom-voice: `check_model_inputs` was rewritten to take `func` as
positional, so `@check_model_inputs()` factory calls fail with
`TypeError: missing 1 required positional argument: 'func'`
- tada-1b: `PretrainedConfig.__init_subclass__` now applies `@dataclass`,
which rejects tada's `strides: list = []` mutable default
- luxtts: Whisper init hits `AssertionError` in `torch._refs.normal_`
Restating the same constraint here keeps mlx-audio's transformers
dependency from quietly winning the resolver.
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
ModelManagement.tsx reads migrationResult.moved (added in #433) but
apiClient.migrateModels() was typed as returning only { source, destination }.
The backend actually returns { moved: int, errors: list[str], source, destination }
(backend/routes/models.py:140, 168). Widen the TS return type so the check
typechecks under the new CI gate from #418.
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
* fix(build): bundle kokoro source files for transformers runtime introspection
transformers opens .py source files at runtime to check attention/MoE
implementation via regex (e.g. _can_set_attn_implementation). PyInstaller's
--hidden-import only bundles .pyc bytecode, so kokoro/modules.py was missing
from the bundle causing a FileNotFoundError on Kokoro model load.
Switch from individual --hidden-import entries to --collect-all kokoro in both
build_binary.py and voicebox-server.spec. The kokoro package is 172K so no
meaningful bundle size impact.
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
* fix(build): use SPECPATH for runtime hook instead of hardcoded absolute path
The linter expanded runtime_hooks=[] to an absolute /Users/... path which
would break CI and other dev machines. Use os.path.join(SPECPATH, ...) to
mirror the relative approach in build_binary.py.
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
* fix(build): runtime hook to work around PyInstaller + Python 3.12 import breakages
Four distinct bundling-specific crashes blocked Kokoro and Qwen CustomVoice
from loading in the frozen binary:
1. torch._dynamo import triggered via class-body decorators
(@torch._dynamo.allow_in_graph on PreTrainedModel,
@torch.compiler.disable in flex_attention) pulls in torch._numpy._ufuncs
which crashes on module load with NameError: name 'name' is not defined.
2. AlbertModel (Kokoro) triggers @auto_docstring -> modeling_auto ->
GenerationMixin -> candidate_generator -> sklearn -> scipy, which hits
the same class of bug in scipy.stats._distn_infrastructure (NameError:
name 'obj' is not defined).
3. AutoModel (Qwen) pulls the same sklearn -> scipy chain directly.
4. librosa (required by most TTS engines) -> scipy.signal -> scipy.stats
hits the _distn_infrastructure crash regardless of the transformers
stubs above.
The root cause of (1) and (4) is that PyInstaller's frozen importer runs
module-level `for X in [<list-comp using dir()>]:` loops with an empty
iterable, leaving the loop variable unbound. Trailing `del obj` / unrelated
references then crash.
Fix: a single runtime hook (pyi_rth_torch_compiler_disable.py) installs:
- sys.modules stubs for torch._dynamo and torch._dynamo.config, plus a
meta-path finder for torch._dynamo.* submodules — voicebox never uses
torch.compile/dynamo for inference, so a permissive no-op stub (callable
as decorator, falsey as predicate, context-manager-safe for
TransformGetItemToIndex) is drop-in safe.
- meta-path finder stubs for transformers.utils.auto_docstring and
transformers.generation.candidate_generator — both import-chain
short-circuits; docstrings and speculative decoding aren't used for TTS.
- meta-path finder for scipy.stats._distn_infrastructure that reads the
real .py source via the wrapped loader's get_source(), replaces the
bundling-broken `del obj` with `globals().pop('obj', None)`, and
compile+exec's the patched source. This keeps the real scipy module
intact so librosa and everything downstream works normally.
Supporting changes:
- backend/pyi_hooks/hook-scipy.stats._distn_infrastructure.py sets
module_collection_mode = "pyz+py" so the .py source is actually in the
bundle for the runtime patcher to read.
- build_binary.py and voicebox-server.spec register the runtime hook and
the new hooks dir.
Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
* fix(build): force transformers torch<2.6 mask path and bundle spacy_pkuseg
- patch transformers.masking_utils to set _is_torch_greater_or_equal_than_2_6
= False, forcing sdpa_mask_older_torch and avoiding the vmap .item() crash
that breaks Qwen CustomVoice generation (our torch._dynamo stub can't
reproduce TransformGetItemToIndex's graph transform).
- add PyInstaller hook to bundle transformers.masking_utils .py source so the
runtime finder can source-patch it.
- --collect-all spacy_pkuseg so Chatterbox Multilingual can load its Chinese
segmenter (dicts/default.pkl + native .so extensions).
- add per-finder install diagnostics + _HOOK_VERSION marker to make future
bundle-only regressions easier to triage.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* fix(build): pass PyInstaller hook paths relative so .spec is portable
Absolute paths ended up in the auto-regenerated voicebox-server.spec
because build_binary.py prefixed every --runtime-hook and
--additional-hooks-dir with str(backend_dir / ...). That broke builds
on any machine whose checkout wasn't at /Users/jamie/... and anyone
invoking pyinstaller voicebox-server.spec directly.
os.chdir(backend_dir) already runs before PyInstaller (same reason
server.py works as a bare filename), so the backend_dir prefix is
unnecessary. Drop it so the generated spec references pyi_hooks/,
pyi_rth_numpy_compat.py, pyi_rth_torch_compiler_disable.py as repo-
relative paths.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
0.3.0 sometimes stored relative media paths with the data-dir name baked in
(e.g. "data/profiles/<uuid>/sample.wav"). resolve_storage_path joined those
directly with _data_dir, producing "<data_dir>/data/profiles/..." — a
spurious double nest that breaks file reads after upgrading to 0.4.0.
The 0.4.0 startup migration didn't catch it because resolve_storage_path
produced the buggy double-nested path, to_storage_path saw "data" at the
first (legitimate) index, and the normalized value matched the stored value
so the row was skipped.
Strip any leading "data/" component before joining. This unblocks runtime
reads and lets _normalize_storage_paths rewrite the affected rows on next
startup — no manual migration needed.
Fixes "No such file or directory: '<data_dir>/data/profiles/...'" and
associated 404s on GET /audio/<id> after upgrading from 0.3.0 to 0.4.0.
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
