search

nesl / asvspoof2019

Our submission to the ASVspoof 2019: Automatic Speaker Verification Spoofing and Countermeasures Challenge
92 stars 40 forks source link

Only two neurons in the output layer? #10

Open jlian2 opened 4 years ago

jlian2 commented 4 years ago

A naive question. To my knowledge, the number of neurals in the last layer should be number of indentities. (e.g. 1000) But there are only 2(bonafida and spoof) for any speakers? What is the reason/superiority of doing this? Thanks!

wangziqi000 commented 4 years ago

Hi, this work is to perform binary classification of whether a speech file is computer-generated or human-spoken. There are only two classes.