Multiple Choice Learning for Efficient Speech Separation with Many Speakers

David Perera, François Derrida, Théo Mariotte +2

Training speech separation models in the supervised setting raises a permutation problem: finding the best assignation between the model predictions and the ground truth separated…