search

HarryVolek / PyTorch_Speaker_Verification

PyTorch implementation of "Generalized End-to-End Loss for Speaker Verification" by Wan, Li et al.
BSD 3-Clause "New" or "Revised" License
575 stars 166 forks source link

what's the result of timit database for uis-rnn? #15

Closed wuqiangch closed 5 years ago

wuqiangch commented 5 years ago

@HarryVolek what's the result of timit database for uis-rnn? My result is very bad.

HarryVolek commented 5 years ago

What was your testing methodology?

Gaurav470 commented 3 years ago

Hi @wuqiangch, @HarryVolek , We are also getting very bad results on our custom data with 5k speakers and have trained our speaker identification model on this data. Then we created embeddings from the above model and trained UIS-RNN model. We then created d-vector embeddings of wave files from the custom speaker identification model made above and tested the prediction on the UIS-RNN model but got only 1 speaker while the audio has multiple speakers. Thanks in advance.