search

FunAudioLLM / CosyVoice

Multi-lingual large voice generation model, providing inference, training and deployment full-stack ability.
https://funaudiollm.github.io/
Apache License 2.0
6.42k stars 691 forks source link

Streaming inference #638

Open OswaldoBornemann opened 1 week ago

OswaldoBornemann commented 1 week ago

So I want to ask, is that only the AR LLM the streaming part, the NAR Flow Matching remains non-streaming?

aluminumbox commented 1 week ago

flow matching is non-streaming, we use chunk inference to simulate streaming, but it is non-streaming inference actually