search

resemble-ai / Resemblyzer

A python package to analyze and compare voices with deep learning
Apache License 2.0
2.67k stars 419 forks source link

Getting different clusters for the same speaker #19

Closed Rahul-Sindhu closed 4 years ago

Rahul-Sindhu commented 4 years ago

Hey,

I tried the projection demo with my own voice recorded files (SpeakerA), and it seems I'm getting overlapping and the clusters are spread far apart, what do you think could be the issue, and how can I go about fixing it?

Screenshot_20191231_130521

Screenshot_20191231_131010

Thanks

CorentinJ commented 4 years ago

Pass min_dist=1 here, it will disable the automatic clustering of UMAP so that you can better visualize how spread your data really is.

Ensure you have at least 4-5 samples so that your embeddings are not regarded as outliers by UMAP.