Repository navigation
Add official Pocket TTS CPU narration and reusable voice cloning - #1289
Merged
Merged
Conversation
Integrate on-demand six-layer language checkpoints, official voices and reusable reference cloning with the existing narration architecture. Fix cancellation of pending download writes and bound long Pocket reference conditioning. Refs #1288
Contributor
|
@jorgepb96, all required checks have passed for the latest revision of this PR. It is ready for maintainer review and merge consideration. Latest revision: |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add Kyutai's official Pocket TTS CPU implementation as an optional narration engine within the existing TTS architecture. The multilingual catalogue groups seven official six-layer language checkpoints, includes their default voices (including Lola), and offers the remaining named official voices on demand. Recorded/uploaded voices can be saved without a transcript and reused across engine restarts through local safetensors conditioning caches.
No Pocket weights or voice states ship in installers. Downloads use pinned revisions, exact sizes and SHA-256 verification, and gated cloning weights use the existing Hugging Face terms/token flow. Pocket runs on macOS arm64 and Windows/Linux x64; it does not use MLX or CUDA.
Live verification also exposed two issues addressed here:
Related issue
Closes #1288
Type of change
Validation
Passed locally:
npm run typechecknode --test scripts/test-tts-download-cancel.mjs scripts/test-tts-gated-download.mjs scripts/test-pocket-tts.mjs scripts/test-pocket-tts-ui.mjs scripts/test-tts-catalogue.mjs scripts/test-tts-host-lifecycle.mjspython -B scripts/test_pocket_tts_backend.pyin the Pocket environment.python runtime/ming-tts/verify_locks.py pocket: verified wheel/hash availability for macOS arm64, Windows x64 and Linux x64.The initial successful cloned narration produced 8.96 seconds of audio in 4.377 seconds of CPU generation. See live verification report and
scripts/verify-pocket-tts-live.mjsfor evidence and reproduction.Full repository lint, test, build, E2E and platform gates are delegated to the PR's GitHub Actions workflows; they are not claimed as completed local runs. The authenticated live campaign remains opt-in and does not require CI credentials.
Screenshots
Automated UI coverage exercises the engine controls, optional voice downloads and saving uploaded audio without a transcript. No personal voice recordings, generated voice samples or private-profile screenshots are attached to this public PR.
Privacy and data review
Contributor checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.md.