performance and loss of the frame-based model

YuanGongND / ssast

Code for the AAAI 2022 paper "SSAST: Self-Supervised Audio Spectrogram Transformer".

BSD 3-Clause "New" or "Revised" License

357 stars 58 forks source link

performance and loss of the frame-based model #32

Closed fabianbosshard closed 2 months ago

fabianbosshard commented 2 months ago

I also wanted to ask if you still have the result.csv from the frame-based base model trained on audioset2m/librispeech, i.e. mask01-base-f128-t2-b24-lr1e-4-m400-pretrain_joint-asli. If you still have it, we would be very thankful if you could upload it here. We are currently trying to reproduce the pretraining results, such that we can then build on that and finetune it for speaker verification.

Best Regards, Fabian

YuanGongND commented 2 months ago

result.csv

It has been a long while, but I tried to find this in our server.

We re-run experiment with cleaned up code and generate these logs, there might be a small difference between this and that reported in the paper.

-Yuan

fabianbosshard commented 2 months ago

Thank you very much for your help! We appreciate the time and effort you've put into this.

-Fabian