Padding-free inference #2813
amineabdaoui
started this conversation in
General
Padding-free inference
#2813
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi @jongwook
Any plan to release a model trained with an encoder attention mask?
This would allow padding-free inference as the current fixed-30s chunks adds a lot of unnecessary computation for short audio inputs.
The speed-up will be huge (c.f. https://ieeexplore.ieee.org/abstract/document/11462351) in this case.
Thanks.
All reactions