Unified Speech-Text Pre-training for Speech Translation and Recognition

TL;DR

This paper introduces a unified speech-text pre-training approach that combines self-supervised and supervised tasks to improve speech translation and recognition, effectively integrating linguistic information from text data.

Contribution

It proposes a novel multi-task pre-training framework that fuses speech and text modeling, addressing subtask interference and enhancing performance on speech translation and recognition.

Findings

Achieves 1.7-2.3 BLEU improvement on MuST-C dataset.

Attains WERs comparable to wav2vec 2.0 on Librispeech.

Effectively integrates linguistic information into speech models.

Abstract

We describe a method to jointly pre-train speech and text in an encoder-decoder modeling framework for speech translation and recognition. The proposed method incorporates four self-supervised and supervised subtasks for cross modality learning. A self-supervised speech subtask leverages unlabelled speech data, and a (self-)supervised text to text subtask makes use of abundant text training data. Two auxiliary supervised speech tasks are included to unify speech and text modeling space. Our contribution lies in integrating linguistic information from the text corpus into the speech pre-training. Detailed analysis reveals learning interference among subtasks. Two pre-training configurations for speech translation and recognition, respectively, are presented to alleviate subtask interference. Our experiments show the proposed method can effectively fuse speech and text information into one model. It achieves between 1.7 and 2.3 BLEU improvement above the state of the art on the MuST-C speech translation dataset and comparable WERs to wav2vec 2.0 on the Librispeech speech recognition task.