Does Joint Training Really Help Cascaded Speech Translation?

TL;DR

This paper investigates whether joint training improves cascaded speech translation, finding that strong baselines diminish its benefits and proposing alternative approaches to leverage direct data more effectively.

Contribution

The study critically evaluates joint training methods in cascaded speech translation and suggests alternative strategies based on experimental analysis.

Findings

Strong cascaded baselines reduce joint training benefits

Joint training offers limited improvements in current setups

Alternative methods may better utilize direct data for speech translation

Abstract

Currently, in speech translation, the straightforward approach - cascading a recognition system with a translation system - delivers state-of-the-art results. However, fundamental challenges such as error propagation from the automatic speech recognition system still remain. To mitigate these problems, recently, people turn their attention to direct data and propose various joint training methods. In this work, we seek to answer the question of whether joint training really helps cascaded speech translation. We review recent papers on the topic and also investigate a joint training criterion by marginalizing the transcription posterior probabilities. Our findings show that a strong cascaded baseline can diminish any improvements obtained using joint training, and we suggest alternatives to joint training. We hope this work can serve as a refresher of the current speech translation landscape, and motivate research in finding more efficient and creative ways to utilize the direct data for speech translation.