v3.2.0
kyutai-labs/pocket-ttsv3.2.0Sep 23, 2026by gabrieldemarmiesse
Upgrading?
AI Summary
This release upgrades the French language support with a new 24-layer model and updates the default English model to a 2026-09 robustness fine-tune. It also introduces new community models for languages like Polish, Welsh, and Estonian, alongside various decoder and training performance improvements.
Key Highlights
- French language updated with a new 24-layer model (`french_24l`).
- Default English model bumped to `english_2026-09` for improved robustness.
- New community models added: Polish, Welsh (Cymraeg), Estonian, and Farsi.
- Decoder thread optimization to decode every queued latent per call.
- Training improvements including Muon optimizer and better checkpointing.
Breaking Changes
- Default English model changed to `english_2026-09`.
- Refusal of manifests without word alignments at startup.
New Features
- New `french_24l` model.
- New community models: Farsi, Indonesian 24L, Estonian, Welsh, and Polish.
- Added `capitalize_first_letter` feature for phoneme models.
- Muon optimizer support for training.
- Length-bucketed batches by default.
- Pocket-TTS-LiteRT (Android GPU) implementation documented.
Full Release Notes
The french language has been updated, we now have `"french"` and a better `"french_24l"`! ## What's Changed * Always end the prompt with sentence-final punctuation by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/296 * README: add Pocket-TTS-LiteRT (Android GPU) to alternative implementations by @john-rocky in https://github.com/kyutai-labs/pocket-tts/pull/295 * Add pocket-tts-farsi to community-trained models by @mallahyari in https://github.com/kyutai-labs/pocket-tts/pull/298 * Add Indonesian 24L community model to README by @anak10thn in https://github.com/kyutai-labs/pocket-tts/pull/294 * Decode every queued latent per call in the decoder thread by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/300 * Add pocket-tts-estonian to community-trained models by @cbentes in https://github.com/kyutai-labs/pocket-tts/pull/301 * README: collapse community model commands into <details> blocks by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/302 * Add dependency cooldown (exclude-newer = 7 days) by @gabrieldemarmiesse in https://github.com/kyutai-labs/pocket-tts/pull/303 * Bump the Indonesian model to its retrained weights by @anak10thn in https://github.com/kyutai-labs/pocket-tts/pull/306 * Add capitalize_first_letter so phoneme models are not capitalized by @mallahyari in https://github.com/kyutai-labs/pocket-tts/pull/307 * Length-bucketed batches by default by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/309 * Add Pocket TTS Cymraeg (Welsh) community model by @dwain-barnes in https://github.com/kyutai-labs/pocket-tts/pull/310 * english_2026-09: robustness fine-tune (mid-word prompt cuts, unpunctuated text) as the default English model by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/311 * Add Pocket TTS Polish (6 layers) community model by @shefowl in https://github.com/kyutai-labs/pocket-tts/pull/315 * Update README.md by @manukyutai in https://github.com/kyutai-labs/pocket-tts/pull/316 * Training: Muon as an optional optimizer by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/314 * Stop generation when nobody is listening anymore by @bfulton in https://github.com/kyutai-labs/pocket-tts/pull/228 * Load tokenizers from tokenizer.json, keeping .model support by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/317 * Refuse manifests without word alignments at startup by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/318 * Ignore EOS on the first 6 generated frames by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/319 * Keep the data loaders alive through SIGTERM so the trainer can checkpoint by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/320 * French: 6-layer model and a new 24-layer model by @manu-lm in https://github.com/kyutai-labs/pocket-tts/pull/321 ## New Contributors * @john-rocky made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/295 * @mallahyari made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/298 * @anak10thn made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/294 * @cbentes made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/301 * @dwain-barnes made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/310 * @shefowl made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/315 * @bfulton made their first contribution in https://github.com/kyutai-labs/pocket-tts/pull/228 **Full Changelog**: https://github.com/kyutai-labs/pocket-tts/compare/v3.1.0...v3.2.0
Related Repositories
public-apis/public-apis★ 491,196
A collective list of free APIs
NousResearch/hermes-agent★ 256,050
The agent that grows with you
yt-dlp/yt-dlp★ 200,229
A feature-rich command-line audio/video downloader
microsoft/markitdown★ 191,546
Python tool for converting files and office documents to Markdown.
Significant-Gravitas/AutoGPT★ 187,699
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.