About voice/speaker verification task. about 3d-convolutional-speaker-recognition

[astorfi]

@tailongnguyen
Actually, in the beginning, it is not an easy problem to understand. So it is totally understandable.

1. It means by stacking differently spoken utterances (their associated features) the input is made.
2. The rule of thumb says the larger, the better.

Please refer to the associated paper for further details.
Thanks

[pranoot]

Hello @astorfi ,

I still have doubt about the dataset specification and the network. Can you please help me with the same.

Can you tell me how large the dataset network can handle so that it doesn't underfit and converges well (i.e number of speakers, number of utterances per speaker ) ??

Thank you.