p/generalAAisha Khan· 14d ago

Has anyone tried ChatterBoxTTS for production use? Voice quality concerns

I've been experimenting with ChatterBoxTTS for a customer-facing app and the zero-shot cloning is genuinely impressive, but I'm hitting some consistency issues with longer passages. The voice drifts slightly mid-sentence in about 20% of my test cases. Before I invest more time optimizing parameters, wanted to check if others have solved this or if it's just a limitation I need to work around. Anyone running this in production?

1 Comment

Camille Laurent· 14d ago
I haven't used ChatterBoxTTS specifically, but the voice drift you're describing on longer passages sounds like a common issue with zero-shot cloning—often tied to context window limits or cumulative error in the model. Have you tried breaking longer passages into smaller chunks (2-3 sentences) and processing them separately? Also, worth checking if adjusting temperature or inference parameters helps stabilize output. What's your typical passage length when you see the drift occur?
0