Adapting Diarization-Conditioned Whisper for End-to-End Multi-Talker Speech Recognition
This addresses the challenge of multi-talker speech recognition for applications like meeting transcription, but it is incremental as it builds on existing Whisper and SOT methods.
The paper tackled the problem of transcribing overlapping speech by proposing a speaker-attributed Whisper-based model that combines target-speaker modeling with serialized output training, resulting in performance that outperforms existing SOT-based approaches and surpasses DiCoW on multi-talker mixtures like LibriMix.
We propose a speaker-attributed (SA) Whisper-based model for multi-talker speech recognition that combines target-speaker modeling with serialized output training (SOT). Our approach leverages a Diarization-Conditioned Whisper (DiCoW) encoder to extract target-speaker embeddings, which are concatenated into a single representation and passed to a shared decoder. This enables the model to transcribe overlapping speech as a serialized output stream with speaker tags and timestamps. In contrast to target-speaker ASR systems such as DiCoW, which decode each speaker separately, our approach performs joint decoding, allowing the decoder to condition on the context of all speakers simultaneously. Experiments show that the model outperforms existing SOT-based approaches and surpasses DiCoW on multi-talker mixtures (e.g., LibriMix).