search

yzyouzhang / ASVspoof2021_AIR

Official implementation of our ASVspoof 2021 paper, "UR Channel-Robust Synthetic Speech Detection System for ASVspoof 2021"
MIT License
52 stars 5 forks source link

Detailed performance of DF submission on different channels and attacks. #19

Closed Betciso closed 2 years ago

Betciso commented 2 years ago

Sir, I have read your article.And i have some questions.i can't understand this Figure 4.How did you get this figure4.The test set has no tags,so how did you know which spoofing attack method was used?

yzyouzhang commented 2 years ago

Thank you for your question and your interest in our paper. The meta labels for the ASVspoof 2021 dataset are not released yet. The numbers in Figures 4&6 and Tables 3-5 are all provided by the organizers. When we submitted our scores for the challenge, the organizers sent back these detailed results to us. I believe they have some backends to produce such results. Please let me know if you have further questions. Thanks.

Betciso commented 2 years ago

Thanks for your answer.I get it .

yzyouzhang commented 2 years ago

If you would like to study the effect of channel variation with meta labels, you can refer to our previous Interspeech paper https://github.com/yzyouzhang/Empirical-Channel-CM. We also released the augmented dataset https://zenodo.org/record/5794671.