Highlights:
6 new models were added, really strong for their sizes! Select them with --language
english_18m_2026-10english_18m_refl_2026-10english_7m_2026-10english_7m_refl_2026-10english_2m_2026-10(experimental)english_2m_refl_2026-10(experimental)
To leverage the speed of those new models we also provide binaries of the pocket-tts runtime, that you can download directly from this Github release page. When you download the executable, it has the same API as the existing CLI of pocket-tts (the python one).
What's Changed
- Add Libratory to projects using Pocket TTS by @subev in #327
- Update README.md by @manukyutai in #328
- README: update xn Rust port links to gradium-ai/xn-ptts by @pratikgx in #330
- Fade in the first 5 ms of each chunk to remove the onset click by @manu-lm in #332
- Fade in the whole first frame of each chunk by @manu-lm in #333
- End voice prompts on a short pause by @manu-lm in #334
- Back to a 5 ms fade-in at chunk starts by @manu-lm in #335
- Drifting sampler head: training recipe and english_drifting_26-09 by @manu-lm in #329
- Add the new languages to the api by @gabrieldemarmiesse in #336
- Add Pocket TTS Greek (Ελληνικά) to community models by @antoniosmakro in #340
- Training: smaller loader defaults and an upfront manifest size check for small finetunes by @manu-lm in #342
- Add ToBe SAID tool information to README by @lookbe in #343
- Retrained French, Italian, Dutch, German, Spanish and Portuguese models (YODAS v3 polish, full-length voice prompts) by @manu-lm in #344
- Add Outmake to projects using Pocket TTS by @DaneBentley in #349
- Limit tests: end repeats, numbers, short texts, bursts, pacing, prompt length, cut prompts, unpunctuated text by @manu-lm in #351
- Add Turkish community model by @alperiox in #352
- Add english_2026-09_2 (re-aligned Dora data) and make it the default by @manu-lm in #353
- Fix cross-page links between the CLI and API reference docs by @pratikgx in #354
- Add Pocket TTS Japanese (日本語) to community models by @takuma104 in #368
- Add Chinese voice option to README by @yuexiajiayan in #373
- Pin predefined voices to the restored pause-ended states (#377) by @manu-lm in #378
- Add multilingual Pocket TTS to community models by @Thanush101 in #381
- Add a rust version of the CLI based on Gradium's xn-ptts by @gabrieldemarmiesse in #383
- Small models: z_dim, attention options, low-rank text table, one-step heads, decoder dims, small English models by @manu-lm in #360
- Rewrite dashes and curly apostrophes the models never saw in training by @quetion667 in #387
- API: serve the six small English models by @gabrieldemarmiesse-ai-agent in #389
- Rust CLI: run the small English models by @gabrieldemarmiesse-ai-agent in #388
New Contributors
- @subev made their first contribution in #327
- @pratikgx made their first contribution in #330
- @antoniosmakro made their first contribution in #340
- @lookbe made their first contribution in #343
- @DaneBentley made their first contribution in #349
- @alperiox made their first contribution in #352
- @takuma104 made their first contribution in #368
- @yuexiajiayan made their first contribution in #373
- @Thanush101 made their first contribution in #381
- @quetion667 made their first contribution in #387
Full Changelog: v3.3.0...v4.0.0