How to Accelerate Audio Generation to Real-Time Speeds
#4
by
samarthshrivas
- opened
Is there a way to speed up the audio generation part to match it with realtime?
Right now on google colab T4 it takes around 28 sec to generate 12 sec audio clip.
@samarthshrivas not sure but I think torch.compile might work? maybe try that? torch.compile(model.text_encoder) and the same for other parts?. You could also use streaming, that might help too?
@samarthshrivas Did you find any way to make it accelerated more and be more efficient in real-time applications?