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 165 forks source link

about the sampling rate #42

Closed nkcdy closed 5 years ago

nkcdy commented 5 years ago

how much does the sampling rate affect the inference? Can I train the network at 16khz sampling rate while inference at other sampling rate such as 32khz or 48khz?

HarryVolek commented 5 years ago

Hi nkcdy,

This question depends on how you preprocess the data before feeding into the NN. I would recommend downsampling the high frequency data into the frequency the network was trained at.