Very different result with sepformer speech enhancement

#1
by Atefeh1987 - opened

I use speecbrain.pretrained library for speech enhancement, sepformer algorithm, but in each run, I get very different result!
Word error rate is from 25 to 75 percent.
How can I handle this problem?

SpeechBrain org
edited Feb 19

Sorry for the very late reply.
On which dataset did you get this? Are you sure that there is no inherent randomness in your dataset?

Sign up or log in to comment