A soft-minimum over speaker-label permutations during training, derived from a probabilistic model, improves SDR and SIR for two-talker speech separation on TIMIT and GRID.
A long-lasting problem in speech separation task is finding the correct label for each separated speech signal, which referred to as label permutation ambiguity
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Probabilistic Permutation Invariant Training for Speech Separation
A soft-minimum over speaker-label permutations during training, derived from a probabilistic model, improves SDR and SIR for two-talker speech separation on TIMIT and GRID.