A Comparative Study on Transformer vs RNN in Speech Applications
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.
Cite
@article{arxiv.1909.06317,
title = {A Comparative Study on Transformer vs RNN in Speech Applications},
author = {Shigeki Karita and Nanxin Chen and Tomoki Hayashi and Takaaki Hori and Hirofumi Inaguma and Ziyan Jiang and Masao Someki and Nelson Enrique Yalta Soplin and Ryuichi Yamamoto and Xiaofei Wang and Shinji Watanabe and Takenori Yoshimura and Wangyou Zhang},
journal= {arXiv preprint arXiv:1909.06317},
year = {2021}
}
Comments
Accepted at ASRU 2019