Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.8.0 #395

2026-08-01T02:01:18Z

Renovate commented

2026-08-01 02:01:18 +00:00

This PR contains the following updates:

Package	Update	Change
ghcr.io/remsky/kokoro-fastapi-cpu	minor	`v0.6.0` → `v0.8.0`

Release Notes

remsky/Kokoro-FastAPI (ghcr.io/remsky/kokoro-fastapi-cpu)

`v0.8.0`

Compare Source

Added

Multi-speaker input on /v1/audio/speech and /dev/captioned_speech (#294). Opt in per request with allow_voice_tags: true; disable server-wide with ENABLE_VOICE_TAGS=false.
- Inline [voice:name] tags switch speaker mid-text.
- voice_aliases mapping for named weighted voice mixes, with optional per-alias rate.
- /dev/captioned_speech timestamps carry the resolved voice per word; the field is absent unless allow_voice_tags is on, so existing responses are unchanged.
POST /dev/dialogue for ordered multi-speaker turns.
SSML input (experimental). Disable server-wide with ENABLE_SSML=false.
- ssml: true on /v1/audio/speech and /dev/captioned_speech translates and speaks in one call. Requires allow_voice_tags: true, since the translation emits [voice:] and [rate:] spans.
- POST /dev/ssml returns the translated tokens as text instead, for inspecting them before synthesis.
return_timing on /v1/audio/speech: per-chunk {text, start, end} JSON sidecar next to the download (powers the web reader).
MAX_PAUSE_DURATION_S (default 60) caps a single [pause:Ns] tag or SSML <break>.
Web UI:
- Voice alias/tag cast builder with import/export, pinning, and per-alias rate, synced with the editor (re: parallel work by @radzrader, #272).
- Read-along mode: sentence highlighting synced to playback, bidirectional click to seek.
- Find/replace across pages, direct page-number entry, download menu (audio / timings / both).
Wiki pages moved into docs/, versioned alongside the code.

Changed

Docker images compile to bytecode at build, ~40% faster startup.
Containers launch uvicorn directly rather than through uv run, which resolves startup permission failures on Unraid and similar hosts.
[rate:] tags scale the speaking voice's alias rate instead of replacing it, so a voice calibrated to 0.8 stays proportionally slower under [rate:1.1]. Matches how SSML engines treat rate.
Speed bounds (0.25 to 4.0) shared across speed fields and SSML.
Unrecognized .env keys warn at startup instead of refusing to boot.
README config table now covers every setting.

Fixed

Long generations swap from the live stream to the finished file as soon as it lands, so the scrubber shows true duration and seeking works mid-run.
Volume control state reconnected to the player.

Removed

Unused ffmpeg from all images (~600MB); audio encoding already runs through PyAV's bundled copy.
Dead pydub dependency.
Unreachable list form of voice from the speech parser and unused VoiceCombineRequest schema.
Legacy Gradio UI (ui/) code cruft; superseded by the web player since ~v0.2.0
Legacy ONNX config compose vars, endpoints e.g /debug/session_pools.
OUTPUT_DIR, OUTPUT_DIR_SIZE_LIMIT_MB, SAMPLE_RATE settings, never read.

`v0.7.2`

Compare Source

Security

fastapi>=0.128.8, starlette>=1.3.1 to close CVE-2025-62727 (quadratic Range header parsing in FileResponse, reachable through the audio download path) (#500).

Changed

CORS allow_credentials now defaults off. Starlette 1.x echoes the caller's origin with allow-credentials: true where 0.47 returned *; nothing here uses cookies or auth, so this keeps the prior behavior.
Docker build cache moved from GHA to the GHCR registry so forks and local builds can pull it, plus uv cache mounts and reordered test-client layers (#501).
response_format docs (correctly) now list aac as supported.

Fixed

FLAC and WAV no longer lose the tail end of the audio; better muxer header patching at finalize (#497, covers #448 and #463). Diagnosis by @Technologicat.

`v0.7.1`

Compare Source

Added

/v1/download/{filename} takes an optional ?name= save-as name (sanitized, stored extension kept) and sets it in Content-Disposition. Omitting it keeps the previous name.
Web UI keyboard navigation and ARIA labeling across header, player controls, and editor.

Changed

Content-Disposition is now built by FileResponse rather than by hand, so the filename comes back quoted (filename="x.mp3") instead of bare. The name itself is unchanged when ?name= is omitted.
Web UI restyle: better use of space, responsive down to slim widths, playbar pinned to the bottom on narrow viewports.
Waveform slowed and softened, made framerate-independent, respects prefers-reduced-motion.
README: AMD GPU (ROCm) troubleshooting, clarified docker-compose comments.

Fixed

Downloads save as {voice}_{timestamp}.{format}, not the temp name (#338). Covers right-click "Save audio as" too, since Content-Disposition outranks the link's download attribute.
Aborted streams no longer surface as playback failures; a user-initiated MEDIA_ERR_ABORTED is told apart from a real error.
Stream-to-file swap settles pending buffer operations instead of leaving the feeder awaiting forever.

`v0.7.0`

Compare Source

Added

AGENTS.md contributor guidelines, plus SKILL.md notes for the API, benchmarks, and web areas.

Changed / Optimizations

Docker images build on Python 3.12 (project floor stays 3.10 for local installs). Rust dropped from the CPU builder.
Runtime dependencies trimmed to remove deprecated imports
bumped requests,python-dotenv, capped transformers<6
Builds now explicitly require BuildKit (default since Docker 23, ~Jan 2023); utilizing COPY --exclude
Model bake reworked to ensure weights land exactly once (whether prexisting or downloaded at build)
ROCm image now bakes the model at build like CPU/GPU (instead of a first-run fetch)
GPU runtime now only uses torch shipped cuDNN/etc via pip wheels (#482). (see table below for size changes)
Transcription benchmark reports split by device; RTF and first-token baselines refreshed.

Compressed image sizes + new bases:

Image	v0.6.0	v0.7.0	Runtime base
cpu	1.66 GB	1.56 GB	`python:3.10-slim` -> `python:3.12-slim`
gpu	6.81 GB	4.68 GB	`cuda:12.6.3-cudnn-runtime` -> `cuda:12.6.3-base`
gpu (cu128)	8.11 GB	5.23 GB	`cuda:12.8.1-cudnn-runtime` -> `cuda:12.8.1-base`
rocm	13.08 GB	13.36 GB	unchanged; model now baked in

Fixed

Model validation checksums against downloaded release artifact to ensure consistency, downloads first to a temp dir to avoid clobbering pre-existing models in the case of network issues/corrupted downloads.
Model validation also rejects any custom files under 100MB to avoids false pass results (e.g. a 9-byte "Not found"), allowing a re-download instead of passing (#301).
.dockerignore Fixed pycache ignore pattern to **/ to ensure nested .pyc/etc stay out of build contexts.
Removed dead pydub imports from the audio services.
_find_file rejects lookups that escape its search roots (../ sequences, absolute paths) to avoid unintentional exposure of files outside the voices/models/web dirs. Symlinks placed inside those dirs resolve as before.
Text normalizer: anchored decimal regex to prevent quadratic backtracking on digit floods, reordered range substitution so NUMBER_PATTERN no longer swallows hyphens meant as range separators, added version-number handling (2.0.1 renders as "two point zero point one" instead of being split).

Configuration

📅 Schedule: (UTC)

Branch creation
- At any time (no schedule defined)
Automerge
- At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

If you want to rebase/retry this PR, check this box

This PR has been generated by Mend Renovate CLI.

This PR contains the following updates: | Package | Update | Change | |---|---|---| | [ghcr.io/remsky/kokoro-fastapi-cpu](https://github.com/remsky/Kokoro-FastAPI) | minor | `v0.6.0` → `v0.8.0` | --- ### Release Notes <details> <summary>remsky/Kokoro-FastAPI (ghcr.io/remsky/kokoro-fastapi-cpu)</summary> ### [`v0.8.0`](https://github.com/remsky/Kokoro-FastAPI/blob/HEAD/CHANGELOG.md#v080---2026-08-14) [Compare Source](https://github.com/remsky/Kokoro-FastAPI/compare/v0.7.2...v0.8.0) ##### Added - Multi-speaker input on `/v1/audio/speech` and `/dev/captioned_speech` ([#294](https://github.com/remsky/Kokoro-FastAPI/issues/294)). Opt in per request with `allow_voice_tags: true`; disable server-wide with `ENABLE_VOICE_TAGS=false`. - Inline `[voice:name]` tags switch speaker mid-text. - `voice_aliases` mapping for named weighted voice mixes, with optional per-alias `rate`. - `/dev/captioned_speech` timestamps carry the resolved `voice` per word; the field is absent unless `allow_voice_tags` is on, so existing responses are unchanged. - `POST /dev/dialogue` for ordered multi-speaker turns. - SSML input (experimental). Disable server-wide with `ENABLE_SSML=false`. - `ssml: true` on `/v1/audio/speech` and `/dev/captioned_speech` translates and speaks in one call. Requires `allow_voice_tags: true`, since the translation emits `[voice:]` and `[rate:]` spans. - `POST /dev/ssml` returns the translated tokens as text instead, for inspecting them before synthesis. - `return_timing` on `/v1/audio/speech`: per-chunk `{text, start, end}` JSON sidecar next to the download (powers the web reader). - `MAX_PAUSE_DURATION_S` (default 60) caps a single `[pause:Ns]` tag or SSML `<break>`. - Web UI: - Voice alias/tag cast builder with import/export, pinning, and per-alias rate, synced with the editor (re: parallel work by [@radzrader](https://github.com/radzrader), [#272](https://github.com/remsky/Kokoro-FastAPI/discussions/272)). - Read-along mode: sentence highlighting synced to playback, bidirectional click to seek. - Find/replace across pages, direct page-number entry, download menu (audio / timings / both). - Wiki pages moved into `docs/`, versioned alongside the code. ##### Changed - Docker images compile to bytecode at build, \~40% faster startup. - Containers launch uvicorn directly rather than through `uv run`, which resolves startup permission failures on Unraid and similar hosts. - `[rate:]` tags scale the speaking voice's alias rate instead of replacing it, so a voice calibrated to 0.8 stays proportionally slower under `[rate:1.1]`. Matches how SSML engines treat rate. - Speed bounds (0.25 to 4.0) shared across speed fields and SSML. - Unrecognized `.env` keys warn at startup instead of refusing to boot. - README config table now covers every setting. ##### Fixed - Long generations swap from the live stream to the finished file as soon as it lands, so the scrubber shows true duration and seeking works mid-run. - Volume control state reconnected to the player. ##### Removed - Unused `ffmpeg` from all images (\~600MB); audio encoding already runs through PyAV's bundled copy. - Dead `pydub` dependency. - Unreachable list form of `voice` from the speech parser and unused `VoiceCombineRequest` schema. - Legacy Gradio UI (`ui/`) code cruft; superseded by the web player since \~v0.2.0 - Legacy ONNX config compose vars, endpoints e.g `/debug/session_pools`. - `OUTPUT_DIR`, `OUTPUT_DIR_SIZE_LIMIT_MB`, `SAMPLE_RATE` settings, never read. ### [`v0.7.2`](https://github.com/remsky/Kokoro-FastAPI/blob/HEAD/CHANGELOG.md#v072-2026-08-06) [Compare Source](https://github.com/remsky/Kokoro-FastAPI/compare/v0.7.1...v0.7.2) ##### Security - `fastapi>=0.128.8`, `starlette>=1.3.1` to close CVE-2025-62727 (quadratic `Range` header parsing in `FileResponse`, reachable through the audio download path) ([#500](https://github.com/remsky/Kokoro-FastAPI/issues/500)). ##### Changed - CORS `allow_credentials` now defaults off. Starlette 1.x echoes the caller's origin with `allow-credentials: true` where 0.47 returned `*`; nothing here uses cookies or auth, so this keeps the prior behavior. - Docker build cache moved from GHA to the GHCR registry so forks and local builds can pull it, plus uv cache mounts and reordered test-client layers ([#501](https://github.com/remsky/Kokoro-FastAPI/issues/501)). - `response_format` docs (correctly) now list `aac` as supported. ##### Fixed - FLAC and WAV no longer lose the tail end of the audio; better muxer header patching at finalize ([#497](https://github.com/remsky/Kokoro-FastAPI/issues/497), covers [#448](https://github.com/remsky/Kokoro-FastAPI/issues/448) and [#463](https://github.com/remsky/Kokoro-FastAPI/issues/463)). Diagnosis by [@Technologicat](https://github.com/Technologicat). ### [`v0.7.1`](https://github.com/remsky/Kokoro-FastAPI/blob/HEAD/CHANGELOG.md#v071---2026-08-02) [Compare Source](https://github.com/remsky/Kokoro-FastAPI/compare/v0.7.0...v0.7.1) ##### Added - `/v1/download/{filename}` takes an optional `?name=` save-as name (sanitized, stored extension kept) and sets it in `Content-Disposition`. Omitting it keeps the previous name. - Web UI keyboard navigation and ARIA labeling across header, player controls, and editor. ##### Changed - `Content-Disposition` is now built by `FileResponse` rather than by hand, so the filename comes back quoted (`filename="x.mp3"`) instead of bare. The name itself is unchanged when `?name=` is omitted. - Web UI restyle: better use of space, responsive down to slim widths, playbar pinned to the bottom on narrow viewports. - Waveform slowed and softened, made framerate-independent, respects `prefers-reduced-motion`. - README: AMD GPU (ROCm) troubleshooting, clarified docker-compose comments. ##### Fixed - Downloads save as `{voice}_{timestamp}.{format}`, not the temp name ([#338](https://github.com/remsky/Kokoro-FastAPI/issues/338)). Covers right-click "Save audio as" too, since `Content-Disposition` outranks the link's `download` attribute. - Aborted streams no longer surface as playback failures; a user-initiated `MEDIA_ERR_ABORTED` is told apart from a real error. - Stream-to-file swap settles pending buffer operations instead of leaving the feeder awaiting forever. ### [`v0.7.0`](https://github.com/remsky/Kokoro-FastAPI/blob/HEAD/CHANGELOG.md#v070---2026-07-31) [Compare Source](https://github.com/remsky/Kokoro-FastAPI/compare/v0.6.0...v0.7.0) ##### Added - `AGENTS.md` contributor guidelines, plus `SKILL.md` notes for the API, benchmarks, and web areas. ##### Changed / Optimizations - Docker images build on Python 3.12 (project floor stays 3.10 for local installs). Rust dropped from the CPU builder. - Runtime dependencies trimmed to remove deprecated imports - bumped `requests`,`python-dotenv`, capped `transformers<6` - Builds now explicitly require BuildKit (default since Docker 23, \~Jan 2023); utilizing `COPY --exclude` - Model bake reworked to ensure weights land exactly once (whether prexisting or downloaded at build) - ROCm image now bakes the model at build like CPU/GPU (instead of a first-run fetch) - GPU runtime now only uses torch shipped cuDNN/etc via pip wheels ([#482](https://github.com/remsky/Kokoro-FastAPI/issues/482)). (see table below for size changes) - Transcription benchmark reports split by device; RTF and first-token baselines refreshed. Compressed image sizes + new bases: <div align="center"> | Image | v0.6.0 | v0.7.0 | Runtime base | | :---------- | -------: | -------: | :------------------------------------------------ | | cpu | 1.66 GB | 1.56 GB | `python:3.10-slim` -> `python:3.12-slim` | | gpu | 6.81 GB | 4.68 GB | `cuda:12.6.3-cudnn-runtime` -> `cuda:12.6.3-base` | | gpu (cu128) | 8.11 GB | 5.23 GB | `cuda:12.8.1-cudnn-runtime` -> `cuda:12.8.1-base` | | rocm | 13.08 GB | 13.36 GB | unchanged; model now baked in | </div> ##### Fixed - Model validation checksums against downloaded release artifact to ensure consistency, downloads first to a temp dir to avoid clobbering pre-existing models in the case of network issues/corrupted downloads. - Model validation also rejects any custom files under 100MB to avoids false pass results (e.g. a 9-byte "Not found"), allowing a re-download instead of passing (#301). - `.dockerignore` Fixed pycache ignore pattern to `**/` to ensure nested .pyc/etc stay out of build contexts. - Removed dead `pydub` imports from the audio services. - `_find_file` rejects lookups that escape its search roots (`../` sequences, absolute paths) to avoid unintentional exposure of files outside the voices/models/web dirs. Symlinks placed inside those dirs resolve as before. - Text normalizer: anchored decimal regex to prevent quadratic backtracking on digit floods, reordered range substitution so `NUMBER_PATTERN` no longer swallows hyphens meant as range separators, added version-number handling (`2.0.1` renders as "two point zero point one" instead of being split). </details> --- ### Configuration 📅 **Schedule**: (UTC) - Branch creation - At any time (no schedule defined) - Automerge - At any time (no schedule defined) 🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied. ♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about this update again. --- - [ ] If you want to rebase/retry this PR, check this box --- This PR has been generated by [Mend Renovate CLI](https://github.com/renovatebot/renovate).

Renovate force-pushed renovate/ghcr.io-remsky-kokoro-fastapi-cpu-0.x from 6dd9ed67f7 to 515862602a

2026-08-03 02:01:31 +00:00

Compare

Renovate changed title from ~~Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.7.0~~ to Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.7.1

2026-08-03 02:01:32 +00:00

Renovate force-pushed renovate/ghcr.io-remsky-kokoro-fastapi-cpu-0.x from 515862602a to 8ce95d8f94

2026-08-07 20:54:46 +00:00

Compare

Renovate changed title from ~~Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.7.1~~ to Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.7.2

2026-08-07 20:54:47 +00:00

Renovate changed title from ~~Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.7.2~~ to Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.8.0

2026-08-16 20:13:27 +00:00

Renovate added 1 commit 2026-08-16 20:13:27 +00:00

Update ghcr.io/remsky/kokoro-fastapi-cpu Docker tag to v0.8.0 96ef9a9925

Renovate force-pushed renovate/ghcr.io-remsky-kokoro-fastapi-cpu-0.x from 8ce95d8f94 to 96ef9a9925

2026-08-16 20:13:27 +00:00

Compare

Lumpiasty merged commit 4af96d3c5d into fresh-start

2026-08-22 21:38:23 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Lumpiasty/klaster#395