Transformer ASR with Contextual Block Processing

Emiru Tsunoo, Yosuke Kashiwagi, Toshiyuki Kumakura, Shinji Watanabe

研究成果: Conference contribution

8 被引用数 (Scopus)

抄録

The Transformer self-Attention network has recently shown promising performance as an alternative to recurrent neural networks (RNNs) in end-To-end (E2E) automatic speech recognition (ASR) systems. However, the Transformer has a drawback in that the entire input sequence is required to compute self-Attention. In this paper, we propose a new block processing method for the Transformer encoder by introducing a context-Aware inheritance mechanism. An additional context embedding vector handed over from the previously processed block helps to encode not only local acoustic information but also global linguistic, channel, and speaker attributes. We introduce a novel mask technique to implement the context inheritance to train the model efficiently. Evaluations of the Wall Street Journal (WSJ), Librispeech, VoxForge Italian, and AISHELL-1 Mandarin speech recognition datasets show that our proposed contextual block processing method outperforms naive block processing consistently. Furthermore, the attention weight tendency of each layer is analyzed to clarify how the added contextual inheritance mechanism models the global information.

本文言語English
ホスト出版物のタイトル2019 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2019 - Proceedings
出版社Institute of Electrical and Electronics Engineers Inc.
ページ427-433
ページ数7
ISBN(電子版)9781728103068
DOI
出版ステータスPublished - 2019 12
外部発表はい
イベント2019 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2019 - Singapore, Singapore
継続期間: 2019 12 152019 12 18

出版物シリーズ

名前2019 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2019 - Proceedings

Conference

Conference2019 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2019
国/地域Singapore
CitySingapore
Period19/12/1519/12/18

ASJC Scopus subject areas

  • コンピュータ ネットワークおよび通信
  • 信号処理
  • 言語学および言語
  • 通信

フィンガープリント

「Transformer ASR with Contextual Block Processing」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル