[Feat]: Please support cosyvoice 3?

[daviddalao]

Description
I downloaded cosyvoice from huggingface and started to chat with it, but it returned meaningless symbols.
Use Case
Describe a scenario where this feature could come handy.

[fgbsgfbb-sys]

This audio was supposed to be generated...
It's unlikely they'll be able to add it anytime soon :(

[a-ghorbani]

CosyVoice is a text-to-speech model — it generates audio tokens, not text. That's why you're seeing meaningless symbols. PocketPal currently only supports GGUF text/chat language models (like Llama, Qwen, Gemma, etc.).

Supporting audio-generating models would require a fundamentally different inference pipeline. It's noted as a request, but not something on the near-term roadmap.