The model targets the decoder part of the system which is the speed bottleneck. ...

srush on Nov 1, 2023 | parent | context | favorite | on: Distil-Whisper: distilled version of Whisper that ...

The model targets the decoder part of the system which is the speed bottleneck. So for tasks like classification it is not likely to be helpful. However a similar method could be used for that use case. (Coauthor)