Speech Recognition and Multi-Speaker Diarization of Long Conversations
This work addresses the challenge of transcribing and labeling speakers in extended conversations like podcasts, which is incremental as it builds on prior joint modeling approaches.
The paper tackled the problem of speech recognition and speaker diarization in long multi-speaker conversations by introducing a new benchmark of hour-long podcasts and finding that joint models outperform separate ones when utterance boundaries are unknown, with improvements from a striding attention decoding algorithm and data augmentation.
Speech recognition (ASR) and speaker diarization (SD) models have traditionally been trained separately to produce rich conversation transcripts with speaker labels. Recent advances have shown that joint ASR and SD models can learn to leverage audio-lexical inter-dependencies to improve word diarization performance. We introduce a new benchmark of hour-long podcasts collected from the weekly This American Life radio program to better compare these approaches when applied to extended multi-speaker conversations. We find that training separate ASR and SD models perform better when utterance boundaries are known but otherwise joint models can perform better. To handle long conversations with unknown utterance boundaries, we introduce a striding attention decoding algorithm and data augmentation techniques which, combined with model pre-training, improves ASR and SD.