Open buaatianye opened 2 years ago
In tools/tdnn.py file, the TDNN class is the ECAPA-TDNN. There are other version of ECAPA-TDNN you can refer: https://github.com/TaoRuijie/ECAPA-TDNN https://github.com/alexw914/ASVspoof_CM/blob/main/models/ecapatdnn.py (This code is provided by speechbrain toolkit)
In tools/tdnn.py file, the TDNN class is the ECAPA-TDNN. There are other version of ECAPA-TDNN you can refer: https://github.com/TaoRuijie/ECAPA-TDNN https://github.com/alexw914/ASVspoof_CM/blob/main/models/ecapatdnn.py (This code is provided by speechbrain toolkit)
thank you for replying, I have found these codes.
by the way, I find one new question, when I training, the result is a little weird, the EER is mostly high and not decreases, is there any setting wrong? 1 16.489909567416294 28.638851536077436 0.7538446260210415 2 11.313554108721538 25.595867900374618 0.650630560969174 3 14.470068523258838 18.010087518551778 0.4936077216257816 4 16.86939028119578 22.73367330668032 0.6256992123250994 5 16.845361762370878 23.737857825767264 0.7336603951813583 6 16.352984097860393 26.493050870763433 0.6653431337198922 7 16.228954693877583 22.268474793792038 0.5898911947015901 8 13.312579350332612 33.181648282197465 0.7677476784126392 9 16.22502685565393 19.868097593159128 0.522340718045202 10 14.381810141998587 25.84368219603975 0.6458102578365725 11 15.647839664255532 20.825543577550786 0.5435535341950974 12 15.512406045024834 21.826346976810846 0.5078309493621691 13 16.110826426570856 27.74504968551551 0.5683558856974903 14 16.382786893381656 35.32406782768458 0.7713560376747267 15 16.889893166069847 24.87947397199432 0.6811595874191827 16 16.481902467857285 31.62157215777742 0.6849971871020873 17 17.861627324113567 35.62200177787188 0.9018587302307058
If you don't apply data augmentation, the EER result will not very well, you must try data augmentation stragegies because ASVspoof2021 introduces the codec transmission problem. You can augment it follow my new project, it provides some other detail things about this research. And it was based on speechbrain toolkit: https://github.com/alexw914/ASVspoof_CM
If you don't apply data augmentation, the EER result will not very well, you must try data augmentation stragegies because ASVspoof2021 introduces the codec transmission problem. You can augment it follow my new project, it provides some other detail things about this research. And it was based on speechbrain toolkit: https://github.com/alexw914/ASVspoof_CM
oh, I see, thanks a lot!
I donnot find the ecapa-tdnn code, is there noly the code of tdnn?