search

TakHemlata / RawBoost-antispoofing

This repository includes the code to reproduce our paper "RawBoost: A Raw Data Boosting and Augmentation Method applied to Automatic Speaker Verification Anti-Spoofing".
MIT License
50 stars 11 forks source link

a question about the equal error rate #1

Closed l1amw closed 2 years ago

l1amw commented 2 years ago

hi, it's a great work, but there is a problem, i used the pre-trained model but found that the EER was about 11.7%, which is quite different from the data in your paper, 5.31%, if i remember correctly, could you please tell me if i did something wrong? Below is the score file. RawBoost_eval_CM_scores.txt

TakHemlata commented 2 years ago

Hi,

Thanks for your interest.

I checked EER performance for your above-attached scores file using performance measure code from ASVspoof 2021 challenge git repo (https://github.com/asvspoof-challenge/2021/tree/main/eval-package/LA/package-stage-1).

Results from your attached scores file: min_tDCF: 0.3099 EER : 5.31 % Which is exactly the same as our paper.

I also again tested my pre-trained model (provided in this repo) on 2021 LA data and got same results as above.

I suggest you before posting please double-check your results with correct codes.

thanks

Regards, Hemlata

TakHemlata commented 2 years ago

I hope you get your answer. I am closing this issue.