Simul-Whisper: Attention-Guided Streaming Whisper with Truncation Detection

TL;DR

Simul-Whisper enables effective streaming speech recognition using Whisper's pre-trained model by leveraging attention-guided decoding and a novel truncation detection method, achieving minimal accuracy loss across languages.

Contribution

This work introduces a novel streaming ASR method that utilizes Whisper's cross-attention for decoding and a new truncation detection model, without requiring fine-tuning.

Findings

Achieves only 1.46% WER degradation at 1s chunks

Outperforms existing streaming ASR baselines

Works across multiple languages and architectures

Abstract

As a robust and large-scale multilingual speech recognition model, Whisper has demonstrated impressive results in many low-resource and out-of-distribution scenarios. However, its encoder-decoder structure hinders its application to streaming speech recognition. In this paper, we introduce Simul-Whisper, which uses the time alignment embedded in Whisper's cross-attention to guide auto-regressive decoding and achieve chunk-based streaming ASR without any fine-tuning of the pre-trained model. Furthermore, we observe the negative effect of the truncated words at the chunk boundaries on the decoding results and propose an integrate-and-fire-based truncation detection model to address this issue. Experiments on multiple languages and Whisper architectures show that Simul-Whisper achieves an average absolute word error rate degradation of only 1.46% at a chunk size of 1 second, which significantly outperforms the current state-of-the-art baseline.