ChatTTS vs VALL-E

ChatTTS is an open-source text-to-speech model built for dialogue. The 2Noise team trained it on over 100,000 hours of Chinese and English speech so it sounds natural in back-and-forth conversation, not just scripted narration.

What sets it apart is prosody control at a granular level. The model can layer in laughter, pauses, and interjections, and it handles multiple speakers in a single session. That makes it a fit for LLM assistants, conversational audio, and dialogue-heavy multimedia.

Developers install it via pip or clone the GitHub repo. The open-source release on Hugging Face is a 40,000-hour base model under AGPLv3+. The team positions it for research and dialogue use cases, with contact at [email protected] for roadmap questions.

VALL-E has developed a context-aware learning function that can be used to synthesize high-quality personalized speech by simply recording an invisible speaker for 3 seconds as a voice prompt. Experimental results show that VALL-E significantly outperforms state-of-the-art zero-shot TTS systems in terms of speech naturalness and speaker similarity. Furthermore, we found that VALL-E can preserve the speaker's emotions and the acoustic environment of the acoustic prompts during synthesis.