search

RookieJunChen / Inter-SubNet

The official PyTorch implementation of "Inter-SubNet: Speech Enhancement with Subband Interaction", accepted by ICASSP 2023.
Apache License 2.0
91 stars 12 forks source link

computationalburden #5

Open shenbuguanni opened 1 year ago

shenbuguanni commented 1 year ago

MACs: 36.71 G The computational power is very large. Has the author tried the performance after the computational power optimization？

RookieJunChen commented 1 year ago

This is because the subbnad units are cut too densely (following what the subband model and FullSubNet do). If you want a smaller amount of computation, you can try relatively less subband units, but this may result in some performance degradation.