Spaces:

sanchit-gandhi

/

musicgen-streaming

like 182

Running on Zero

Just thought I'd mention it would be cool to have something like this for transformers based tts audio streaming. As far as I know, none of the open source models or their libraries currently have the ability to stream audio as its generated, and most of the high quality models take somewhere from 0.5s to 1s to generate 1s of audio, which basically means its not realtime if its not streamed. Having something like this would probably be pretty nice for the whole of open source tts.
*edit hm nevermind I guess xtts might already have a streaming method.