Graph-PIT: Generalized permutation invariant training for continuous separation of arbitrary numbers of speakers.
Thilo von NeumannKeisuke KinoshitaChristoph BöddekerMarc DelcroixReinhold Haeb-UmbachPublished in: CoRR (2021)
Keyphrases
- training set
- supervised learning
- graph representation
- graph theory
- connected components
- graph matching
- graph structure
- speech recognition
- affine transformation
- test set
- training samples
- training process
- directed graph
- object recognition
- directed acyclic graph
- graph model
- training phase
- random walk
- relational structures
- discrete valued
- connected graphs