A Comparative Study on Transformer vs RNN in Speech Applications

TL;DR

This study compares Transformer and RNN models across multiple speech processing tasks, demonstrating Transformer’s superior performance in most benchmarks and providing practical training insights.

Contribution

It offers a comprehensive experimental comparison of Transformer and RNN models in speech applications, with new training tips and open-source recipes.

Findings

Transformer outperforms RNN in 13/15 ASR benchmarks

Provides training tips for Transformer models

Releases reproducible recipes for speech tasks

Abstract

Sequence-to-sequence models have been widely used in end-to-end speech processing, for example, automatic speech recognition (ASR), speech translation (ST), and text-to-speech (TTS). This paper focuses on an emergent sequence-to-sequence model called Transformer, which achieves state-of-the-art performance in neural machine translation and other natural language processing applications. We undertook intensive studies in which we experimentally compared and analyzed Transformer and conventional recurrent neural networks (RNN) in a total of 15 ASR, one multilingual ASR, one ST, and two TTS benchmarks. Our experiments revealed various training tips and significant performance benefits obtained with Transformer for each task including the surprising superiority of Transformer in 13/15 ASR benchmarks in comparison with RNN. We are preparing to release Kaldi-style reproducible recipes using open source and publicly available datasets for all the ASR, ST, and TTS tasks for the community to succeed our exciting outcomes.