Streaming Decoder-Only Automatic Speech Recognition with Discrete Speech Units: A Pilot Study

TL;DR

This paper presents a novel streaming decoder-only speech recognition model that uses discrete speech units, boundary tokens, and advanced attention mechanisms, achieving competitive results on AISHELL datasets.

Contribution

The study introduces a streaming-capable decoder-only ASR model with boundary tokens and right-chunk attention, tailored for real-time speech recognition tasks.

Findings

Achieves competitive performance with non-streaming models on AISHELL datasets.

Employs boundary tokens and right-chunk attention to enhance streaming recognition.

Utilizes data augmentation techniques to improve contextual modeling.

Abstract

Unified speech-text models like SpeechGPT, VioLA, and AudioPaLM have shown impressive performance across various speech-related tasks, especially in Automatic Speech Recognition (ASR). These models typically adopt a unified method to model discrete speech and text tokens, followed by training a decoder-only transformer. However, they are all designed for non-streaming ASR tasks, where the entire speech utterance is needed during decoding. Hence, we introduce a decoder-only model exclusively designed for streaming recognition, incorporating a dedicated boundary token to facilitate streaming recognition and employing causal attention masking during the training phase. Furthermore, we introduce right-chunk attention and various data augmentation techniques to improve the model's contextual modeling abilities. While achieving streaming speech recognition, experiments on the AISHELL-1 and -2 datasets demonstrate the competitive performance of our streaming approach with non-streaming decoder-only counterparts.