sanoTTS is an open-source family of compact neural text-to-speech models and runtimes maintained by Ampixa Labs. It supports local speech synthesis through Python, WebAssembly and embedded C, including selected ESP32-S3 configurations. Its aim is to make speech interfaces practical on inexpensive hardware and in offline or privacy-sensitive applications. The source includes inference code, language bindings, training recipes, evaluation tools and runnable examples. The inference runtime and designated bindings are MIT-licensed; the wider project and espeak-ng integration are GPL-3.0-or-later.
Public adoption evidence checked on 20 November 2026: Hugging Face reports 590,604 downloads in the last month and 1,354,798 all-time for ampixa/sanoTTS; GitHub reports 851 stars and 90 forks. Downloads are platform-counted file requests, not unique people or deployed installations. Independent reuse includes kokopop, which supports sanoTTS inference, and sanoTTS-jp, an independently maintained Japanese reimplementation of the published recipe. These projects retain their own ownership and licences.
Evidence: https://huggingface.co/ampixa/sanoTTS ; https://github.com/Ampixa/sanoTTS ; https://github.com/tterrasson/kokopop ; https://github.com/ayutaz/sanoTTS-jp .
Funding will sustain the existing public software: maintenance, reproducible evaluation, compatibility testing and contributor support. Accessibility is an intended application; the new Android engine remains pre-alpha and user evaluation is still needed.
On 20 November 2026, the Open TTS Leaderboard's CPU streaming table ranks sanoTTS heart-nano #1 (5.9 ms), heart #3 (9.8 ms) and amy #5 (36.3 ms) by p95 time to first audio: three of the top five places. These are latency results; the entries note host-CPU/NumPy evaluation. Source: https://huggingface.co/spaces/hf-audio/open_tts_leaderboard?streaming-cpu
Public adoption evidence checked on 20 November 2026: Hugging Face reports 590,604 downloads in the last month and 1,354,798 all-time for ampixa/sanoTTS; GitHub reports 851 stars and 90 forks. Downloads are platform-counted file requests, not unique people or deployed installations. Independent reuse includes kokopop, which supports sanoTTS inference, and sanoTTS-jp, an independently maintained Japanese reimplementation of the published recipe. These projects retain their own ownership and licences.
Evidence: https://huggingface.co/ampixa/sanoTTS ; https://github.com/Ampixa/sanoTTS ; https://github.com/tterrasson/kokopop ; https://github.com/ayutaz/sanoTTS-jp .
Funding will sustain the existing public software: maintenance, reproducible evaluation, compatibility testing and contributor support. Accessibility is an intended application; the new Android engine remains pre-alpha and user evaluation is still needed.
On 20 November 2026, the Open TTS Leaderboard's CPU streaming table ranks sanoTTS heart-nano #1 (5.9 ms), heart #3 (9.8 ms) and amy #5 (36.3 ms) by p95 time to first audio: three of the top five places. These are latency results; the entries note host-CPU/NumPy evaluation. Source: https://huggingface.co/spaces/hf-audio/open_tts_leaderboard?streaming-cpu