Our ICASSP paper studying permutation ambiguity on speaker-independent source separation models is now accessible on arXiv:
https://arxiv.org/pdf/2102.04945
Preprint: On permutation invariant training for speech source separation
Month: February 2021