Compatibility with pyannote 3.1

juanmc2005 / diart

A python package to build AI-powered real-time audio applications

https://diart.readthedocs.io

MIT License

903 stars 76 forks source link

Compatibility with pyannote 3.1 #212

Closed juanmc2005 closed 7 months ago

juanmc2005 commented 8 months ago

Some things to fully take advantage of pyannote.audio 3.1 (and also verify that nothing's broken):

[x] Verify that WeSpeaker embeddings work as expected
[x] Verify that other WeSpeaker models are also compatible
[x] #214

juanmc2005 commented 7 months ago

Currently working well with pyannote.audio 3.1, a small interpolation change in embedding extraction is changing the results of the paper for AMI, so we probably need to re-tune hyper-parameters