Empowering Whisper as a Joint Multi-Talker and Target-Talker Speech Recognition System

TL;DR

This paper introduces a novel method to adapt the Whisper speech model for simultaneous multi-talker and target-talker speech recognition, achieving superior results on multiple datasets by integrating a separator, target identifier, and soft prompt tuning.

Contribution

It presents a pioneering approach to enable Whisper to perform joint multi-talker and target-talker recognition, combining several techniques for improved performance.

Findings

Outperforms previous methods on LibriMix and LibriSpeechMix datasets.

Achieves acceptable zero-shot performance on AishellMix Mandarin dataset.

Demonstrates the effectiveness of the combined approach for complex speech recognition tasks.

Abstract

Multi-talker speech recognition and target-talker speech recognition, both involve transcription in multi-talker contexts, remain significant challenges. However, existing methods rarely attempt to simultaneously address both tasks. In this study, we propose a pioneering approach to empower Whisper, which is a speech foundation model, to tackle joint multi-talker and target-talker speech recognition tasks. Specifically, (i) we freeze Whisper and plug a Sidecar separator into its encoder to separate mixed embedding for multiple talkers; (ii) a Target Talker Identifier is introduced to identify the embedding flow of the target talker on the fly, requiring only three-second enrollment speech as a cue; (iii) soft prompt tuning for decoder is explored for better task adaptation. Our method outperforms previous methods on two- and three-talker LibriMix and LibriSpeechMix datasets for both tasks, and delivers acceptable zero-shot performance on multi-talker ASR on AishellMix Mandarin dataset.