CJST: CTC Compressor based Joint Speech and Text Training for Decoder-Only ASR

TL;DR

This paper introduces CJST, a novel joint speech and text training framework using CTC compressor for decoder-only ASR, improving text injection and robustness across various conditions.

Contribution

The paper proposes a new CJST framework that leverages CTC compressor features for effective joint speech and text training in decoder-only ASR models.

Findings

Achieves state-of-the-art performance on Librispeech and TED-LIUM2.

Effectively injects text without duration handling.

Provides a comprehensive analysis of CTC compressor behavior.

Abstract

CTC compressor can be an effective approach to integrate audio encoders to decoder-only models, which has gained growing interest for different speech applications. In this work, we propose a novel CTC compressor based joint speech and text training (CJST) framework for decoder-only ASR. CJST matches speech and text modalities from both directions by exploring a simple modality adaptor and several features of the CTC compressor, including sequence compression, on-the-fly forced peaky alignment and CTC class embeddings. Experimental results on the Librispeech and TED-LIUM2 corpora show that the proposed CJST achieves an effective text injection without the need of duration handling, leading to the best performance for both in-domain and cross-domain scenarios. We also provide a comprehensive study on CTC compressor, covering various compression modes, edge case handling and behavior under both clean and noisy data conditions, which reveals the most robust setting to use CTC compressor for decoder-only models.