Self-Taught Recognizer: Toward Unsupervised Adaptation for Speech Foundation Models

TL;DR

This paper introduces STAR, an unsupervised adaptation framework that improves speech recognition models' robustness across diverse domains using unlabeled data, novel quality indicators, and effective model updates.

Contribution

STAR is a novel unsupervised adaptation method that leverages a new token-level quality indicator to enhance speech recognition models without labeled data.

Findings

Achieves 13.5% relative WER reduction across 14 domains

Prevents catastrophic forgetting during adaptation

Requires less than one hour of unlabeled data for effective adaptation

Abstract

We propose an unsupervised adaptation framework, Self-TAught Recognizer (STAR), which leverages unlabeled data to enhance the robustness of automatic speech recognition (ASR) systems in diverse target domains, such as noise and accents. STAR is developed for prevalent speech foundation models based on Transformer-related architecture with auto-regressive decoding (e.g., Whisper, Canary). Specifically, we propose a novel indicator that empirically integrates step-wise information during decoding to assess the token-level quality of pseudo labels without ground truth, thereby guiding model updates for effective unsupervised adaptation. Experimental results show that STAR achieves an average of 13.5% relative reduction in word error rate across 14 target domains, and it sometimes even approaches the upper-bound performance of supervised adaptation. Surprisingly, we also observe that STAR prevents the adapted model from the common catastrophic forgetting problem without recalling source-domain data. Furthermore, STAR exhibits high data efficiency that only requires less than one-hour unlabeled data, and seamless generality to alternative large speech models and speech translation tasks. Our code aims to open source to the research communities.