Access speaker embeddings ?

[ArnaudWald]

Hello, first of all thanks for putting this script together, it works pretty smoothly and accuracy is great ❤️

I want to use this to make speaker recognition across multiple files, i.e. is the speaker_0 in one file the same as speaker_1 in another.
For this I'm thinking about doing some clustering on the speaker embeddings and having a decision threshold based on their similarity.

Is there a step in this script where I can access the speaker embeddings ? I have ran the code step by step but i can't really find this.

Thanks !

[MahmoudAshraf97]

the line that runs diarization runs VAD, Embedding model and clustering internally, you can access the intermediate results in temp_outputs folder or use a separate embedding model from Nemo