search

clovaai / voxceleb_trainer

In defence of metric learning for speaker recognition
MIT License
1.03k stars 272 forks source link

Why using different input features #47

Closed XinhaoMei closed 4 years ago

XinhaoMei commented 4 years ago

Hello, just want to enquire for why using different input features for different network (Mel Spectrogram for VGG, Spectrogram for ResNet)? And which is better? I have tried to use Spectrogram in VGG, and in my experiment, there is not much difference regarding the final performance comparing to Mel Spectrogram.

joonson commented 4 years ago

In our experience the choice of input features doesn't make much difference to performance.