Graph-PIT: Generalized permutation invariant training for continuous separation of arbitrary numbers of speakers
This addresses the challenge of overlapped speech in meeting transcription, enabling more diverse scenarios, though it is an incremental improvement over existing methods.
The paper tackles the problem of continuous speech separation for meeting transcription by relaxing the constraint that the number of speakers must not exceed output channels, proposing a graph-based PIT criterion that allows processing arbitrary numbers of speakers and longer segments, resulting in improved recognition performance on WSJ data.
Automatic transcription of meetings requires handling of overlapped speech, which calls for continuous speech separation (CSS) systems. The uPIT criterion was proposed for utterance-level separation with neural networks and introduces the constraint that the total number of speakers must not exceed the number of output channels. When processing meeting-like data in a segment-wise manner, i.e., by separating overlapping segments independently and stitching adjacent segments to continuous output streams, this constraint has to be fulfilled for any segment. In this contribution, we show that this constraint can be significantly relaxed. We propose a novel graph-based PIT criterion, which casts the assignment of utterances to output channels in a graph coloring problem. It only requires that the number of concurrently active speakers must not exceed the number of output channels. As a consequence, the system can process an arbitrary number of speakers and arbitrarily long segments and thus can handle more diverse scenarios. Further, the stitching algorithm for obtaining a consistent output order in neighboring segments is of less importance and can even be eliminated completely, not the least reducing the computational effort. Experiments on meeting-style WSJ data show improvements in recognition performance over using the uPIT criterion.