search

jymsuper / SpeakerRecognition_tutorial

Simple d-vector based Speaker Recognition (verification and identification) using Pytorch
MIT License
210 stars 46 forks source link

enroll_per_spk does not average #11

Closed cirograciapons closed 3 years ago

cirograciapons commented 3 years ago

https://github.com/jymsuper/SpeakerRecognition_tutorial/blob/6dce64657632660a791e25624e4257c53c7257d6/enroll.py#L73 states that outputs the averaged d-vector

however there is not averagin operation and only agregates embeddings https://github.com/jymsuper/SpeakerRecognition_tutorial/blob/6dce64657632660a791e25624e4257c53c7257d6/enroll.py#L90

is that true?

jymsuper commented 3 years ago

You are right. To average d-vectors, you have to divide the summation by the number of files. However, in this case, there is only one file. Therefore, it does not matter.