Music Embedding Clustering Using a Pretrained Speaker Verification Model. Can a model trained for speaker verification separate songs from different bands?
Given a dataset with songs from some artists, I have extracted 15s excerpts from these songs and generated an embedding with ECAPA-TDNN pretrained for speaker-verification task on the VoxCeleb2 dataset.
Once we have the embeddings, we can visualize them on a TSNE plot:
The artists where the vocal components are the most predominant, like pop and rap, are the ones that the model is capable to separate the best.
Interestingly, the techno genre represented by Boris Brejcha is also nicely separated and is closer to the metal and rock bands than to rap and pop
I intend to come back at this task to finetune the model for genre/artist/album identification.