search

mhlevgen / DCUNetTorchSound

Implementation of Phase-aware speech enhancement with deep complex U-Net
38 stars 8 forks source link

PESQ computation takes long time #2

Open jonashaag opened 4 years ago

jonashaag commented 4 years ago

In my setup, computing PESQ of all batches takes ~50% of total training time (with DCUNet-20).

Maybe we can compute PESQ in the background (it's on CPU anyways), or compute PESQ of a subset only.

For my experiments I've changed the code to only compute PESQ of 10% of samples.

mhlevgen commented 4 years ago

When I trained, I didn't check time for PESQ explicitly, because it wasn't suspicious. I think 10% in training time is fine. Then after training, it's better to compute on whole data.