search

YuanGongND / psla

Code for the TASLP paper "PSLA: Improving Audio Tagging With Pretraining, Sampling, Labeling, and Aggregation".
BSD 3-Clause "New" or "Revised" License
140 stars 16 forks source link

Why you choose to use fbank instead of spectrogram? #9

Open nanyyyyyy opened 1 year ago

nanyyyyyy commented 1 year ago

Why you choose to use fbank instead of spectrogram? Thank you!

YuanGongND commented 1 year ago

Hi there,

I don't have a strong reason to choose one over another. But I felt that fbank is more standard for audio application as it provides some flexibility on the choice of filters (in this repo, mel filters). Having said that, I haven't formally studied if mel filters are better for general audio classification. I would expect the normalization stats for spectrogram would be different with fbanks, so if you want to choose to use spectrogram, please double check the normalization mean and std (especially for ImageNet pretrained models).

-Yuan