search

LAION-AI / natural_voice_assistant

MIT License
431 stars 35 forks source link

StyleTTS2 for german language #21

Open ichDaheim opened 4 months ago

ichDaheim commented 4 months ago

A version of StyleTTS2 generating german language with good emphasis would be a giant leap forward in my opinion. All the existing open source TTS systems for german are sounding somewhat robotic or indifferent. (no comparison to the english versions - as far as i can judge as a non native english speaking person).

ichDaheim commented 4 months ago

Not sure because of the Licenses but maybe these are helpful ? Mozilla Common Voice: (approximately 1300 hours of german audio) https://commonvoice.mozilla.org/de/datasets

Thorsten-Voice: https://github.com/thorstenMueller/Thorsten-Voice?tab=readme-ov-file#voice-datasets

braunale commented 4 months ago

Hey @ichDaheim, thanks for your post. You are right, mulit-language support is definitely a very important point and already part of our roadmap! StyleTTS2, which we are currently using as TTS system, unfortunately only supports english. We experimented with other TTS models which also supports other languages, but all of them either have a bad voice quality or are significantly slower than StyleTTS. We are currently working on a streamable TTS system than can start synthesizing with just a few text tokens look-ahead. This could drastically decrease the latency and we could then fine-tune it on other languages.

If you are interessted in contributing to this or know any open-source TTS that supports (token-by-token) streaming, I am very happy to chat about it in detail.