Skip to main navigation Skip to search Skip to main content

Regularized Forward-Backward Decoder for Attention Models

  • Technical University of Munich

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Nowadays, attention models are one of the popular candidates for speech recognition. So far, many studies mainly focus on the encoder structure or the attention module to enhance the performance of these models. However, mostly ignore the decoder. In this paper, we propose a novel regularization technique incorporating a second decoder during the training phase. This decoder is optimized on time-reversed target labels beforehand and supports the standard decoder during training by adding knowledge from future context. Since it is only added during training, we are not changing the basic structure of the network or adding complexity during decoding. We evaluate our approach on the smaller TEDLIUMv2 and the larger LibriSpeech dataset, achieving consistent improvements on both of them.

Original languageEnglish
Title of host publicationSpeech and Computer - 23rd International Conference, SPECOM 2021, Proceedings
EditorsAlexey Karpov, Rodmonga Potapova
PublisherSpringer Science and Business Media Deutschland GmbH
Pages786-794
Number of pages9
ISBN (Print)9783030878016
DOIs
StatePublished - 2021
Event23rd International Conference on Speech and Computer, SPECOM 2021 - Virtual, Online
Duration: 27 Sep 202130 Sep 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12997 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference23rd International Conference on Speech and Computer, SPECOM 2021
CityVirtual, Online
Period27/09/2130/09/21

Keywords

  • Attention models
  • Forward-backward decoder
  • Regularization
  • Speech recognition

Fingerprint

Dive into the research topics of 'Regularized Forward-Backward Decoder for Attention Models'. Together they form a unique fingerprint.

Cite this