Attempt the following to speed up TTS:
- For long texts, when submitting to TTS, break them down into chunks (e.g., ~100 tokens) to avoid decoder crashes and improve speed. Do not break up sentences (you can go over 100 tokens if needed).
- Experiment with TinyTTS, Piper, Kokoro and KittenTTS using sherpa-onnx to see if either is particularly faster than existing solution. TinyTTS is so small we may be able to embed it in-app as a default rather than downloading larger models (easier distribution, use offline).
- Potentially try to use GPU to speed up TTS decoding (and/or voice input!)
- When sending AI text to the text-to-speech mode, remove markdown formatting. Currently the tts will speak "asterisk asterisk hello asterisk asterisk" for the text
**hello**, which is quite annoying
Attempt the following to speed up TTS:
- For long texts, when submitting to TTS, break them down into chunks (e.g., ~100 tokens) to avoid decoder crashes and improve speed. Do not break up sentences (you can go over 100 tokens if needed).
- Experiment with TinyTTS, Piper, Kokoro and KittenTTS using sherpa-onnx to see if either is particularly faster than existing solution. TinyTTS is so small we may be able to embed it in-app as a default rather than downloading larger models (easier distribution, use offline).
- Potentially try to use GPU to speed up TTS decoding (and/or voice input!)
- When sending AI text to the text-to-speech mode, remove markdown formatting. Currently the tts will speak "asterisk asterisk hello asterisk asterisk" for the text `**hello**`, which is quite annoying