About streaming inference #15

Activity

1. shenduldh commented on Apr 30, 2026

   

   shenduldh

   on Apr 30, 2026

   Author

   More actions

   I have read the paper, maybe 0.5s?
2. aask1357 commented on Apr 30, 2026

   

   aask1357

   on Apr 30, 2026

   Owner

   More actions

   For each model, minimum chunk size is same as the hop size. So for Tiny, Base, Small, their chunk sizes are 256/16000 seconds = 16ms. For Medium, it's 160/16000 = 1ms. For large, it's 100/16000 = 6.25ms.
3. shenduldh commented on May 8, 2026

   

   shenduldh

   on May 8, 2026

   Author

   More actions

   Thanks. But I use the min chunk size to denoise speech will cause strange noise.
4. aask1357 commented on May 11, 2026

   

   aask1357

   on May 11, 2026

   Owner

   More actions

   That should be a bug. Can you tell me more in detail about the model and configuration you used?