taylorlu / Speaker-Diarization

speaker diarization by uis-rnn and speaker embedding by vgg-speaker-recognition

Apache License 2.0

455 stars 124 forks source link

File size exceed Zip #16

Open Arroosh opened 4 years ago

Arroosh commented 4 years ago

Hello everyone Screenshot from 2019-07-09 19-30-43

I set the range of epoch 4000 in generate_embeddings.py but i get the following error for epoch in range(4000): # Random choice utterances from whole wavfiles

A merged utterance contains [10,20] utterances

Please guide me how to set range? Thanks in advance

Arroosh commented 4 years ago

The range of epoch (i.e in my case i set it 4000) only shows that the model will be exposed to or pass through the whole dataset 4000 times. I think i am missing something very important??? @taylorlu @giorgionanfa please help me to resolve this issue. Thanks in advance

Arroosh commented 4 years ago

Screenshot from 2019-07-10 14-16-03 I have created embeddings after setting epoch= 2000 but i'm unable to train it using CPU mode. "killed" message appears after running command "python train.py" please let me know if you find any clue from this behaviour. Thanks in advance

chrisspen commented 4 years ago

Looks like you ran into this issue. By default, the uis-rnn train() method loads all training data into memory, so if you have more data than memory, your OS will possibly kill your process. They claim you can work around this splitting up your data and calling fit() on each part.