search

zycv / OpenSpeaker

OpenSpeaker is a completely independent and open source speaker recognition project. It provides the entire process of speaker recognition including multi-platform deployment and model optimization.
Apache License 2.0
60 stars 10 forks source link

two different audio files get cosine similarity over 0.9 #7

Open YunzhaoLu opened 2 years ago

YunzhaoLu commented 2 years ago

Almost most of different audio files get cosine similarity over 0.9 ? How to interpret the similarity value?

zycv commented 2 years ago

Hi, YunzhaoLu, the threshold of the model is determined according to the EER of the test set. The model here is just a demo. It is recommended that you can train one yourself.

YunzhaoLu commented 2 years ago

When I fix a test set, I get the threshold. But how can I apply the threshold to an unseen dataset?