search

eurecom-asp / rawnet2-antispoofing

This repository includes the code to reproduce our paper "End-to-end anti-spoofing with RawNet2" (https://arxiv.org/abs/2011.01108) published in ICASSP '21.
MIT License
48 stars 13 forks source link

some problems about this paper result #4

Closed ZigHello closed 1 year ago

ZigHello commented 1 year ago

Dear author: Thank you so much for such a great job，I recently reproduced your work on the ASVspoof2019 LA dataset, I used s1, s2, s3 pre-trained models and did not get the results in the article,the best result is from the s1 pre-trained model, eer: 20.62%, which is a big gap with the results in the article, did I miss something?

image image
TakHemlata commented 1 year ago

Hello ZigHello,

Thanks for your interest. I again checked with pre-trained model and I got similar result. I assume maybe you did some mistake in EER computation. Make sure that your score file is matched with protocol file. Otherwise it will give wrong performance.

Thanks

TakHemlata commented 1 year ago

Hi Bonlin0

Thanks for letting me know. The pre-trained model was by mistake wrong. I suggest you check this update github repo "https://github.com/asvspoof-challenge/2021/tree/main/LA/Baseline-RawNet2" and train from the start or check with pre-trained model. This one we used for the 2021 challenge but the pre-trained model is on 2019 LA training data only.

I hope that helps. Thanks