End-to-end speaker segmentation for overlap-aware resegmentation
This work addresses speaker segmentation for conversations, offering a more efficient approach by eliminating the need for combining multiple sub-tasks, though it is incremental as it builds on existing end-to-end neural diarization methods.
The paper tackled speaker segmentation by proposing an end-to-end model that directly partitions conversations into speaker turns, achieving relative diarization error rate improvements of 17% on AMI, 13% on DIHARD 3, and 13% on VoxConverse over the best baseline.
Speaker segmentation consists in partitioning a conversation between one or more speakers into speaker turns. Usually addressed as the late combination of three sub-tasks (voice activity detection, speaker change detection, and overlapped speech detection), we propose to train an end-to-end segmentation model that does it directly. Inspired by the original end-to-end neural speaker diarization approach (EEND), the task is modeled as a multi-label classification problem using permutation-invariant training. The main difference is that our model operates on short audio chunks (5 seconds) but at a much higher temporal resolution (every 16ms). Experiments on multiple speaker diarization datasets conclude that our model can be used with great success on both voice activity detection and overlapped speech detection. Our proposed model can also be used as a post-processing step, to detect and correctly assign overlapped speech regions. Relative diarization error rate improvement over the best considered baseline (VBx) reaches 17% on AMI, 13% on DIHARD 3, and 13% on VoxConverse.