Stereo-input speech recognition using sparseness-based time-frequency masking in a reverberant environment

Yosuke Izumi, Kenta Nishiki, Shinji Watanabe, Takuya Nishimoto, Nobutaka Ono, Shigeki Sagayama

Research output: Contribution to journalArticle

Abstract

We present noise robust automatic speech recognition (ASR) using sparseness-based underdetermined blind source separation (BSS) technique. As a representative underdetermined BSS method, we utilized time-frequency masking in this paper. Although time-frequency masking is able to separate target speech from interferences effectively, one should consider two problems. One is that masking does not work well in noisy or reverberant environment. Another is that masking itself might cause some distortion of the target speech. For the former, we apply our time-frequency masking method [7] which can separate the target signal robustly even in noisy and reverberant environment. Next, investigating the distortion caused by time-frequency masking, we reveal following facts through experiments: 1) soft mask is better than binary mask in terms of recognition performance and 2) cepstral mean normalization (CMN) reduces the distortion, especially for that caused by soft mask. At the end, we evaluate the recognition performance of our method in noisy and reverberant real environment.

Original languageEnglish
Pages (from-to)1955-1958
Number of pages4
JournalUnknown Journal
Publication statusPublished - 2009
Externally publishedYes

Fingerprint

Speech intelligibility
speech recognition
masking
Speech recognition
Masks
Blind source separation
masks
Acoustic noise
Noise
Experiments
interference
causes

Keywords

  • Blind source separation
  • Robust ASR
  • Speech sparseness
  • Stereo-input
  • Time-frequency mask

ASJC Scopus subject areas

  • Human-Computer Interaction
  • Signal Processing
  • Software
  • Sensory Systems

Cite this

Stereo-input speech recognition using sparseness-based time-frequency masking in a reverberant environment. / Izumi, Yosuke; Nishiki, Kenta; Watanabe, Shinji; Nishimoto, Takuya; Ono, Nobutaka; Sagayama, Shigeki.

In: Unknown Journal, 2009, p. 1955-1958.

Research output: Contribution to journalArticle

Izumi, Yosuke ; Nishiki, Kenta ; Watanabe, Shinji ; Nishimoto, Takuya ; Ono, Nobutaka ; Sagayama, Shigeki. / Stereo-input speech recognition using sparseness-based time-frequency masking in a reverberant environment. In: Unknown Journal. 2009 ; pp. 1955-1958.
@article{b72a1f34e7284ca2a3340f2db0be6ee6,
title = "Stereo-input speech recognition using sparseness-based time-frequency masking in a reverberant environment",
abstract = "We present noise robust automatic speech recognition (ASR) using sparseness-based underdetermined blind source separation (BSS) technique. As a representative underdetermined BSS method, we utilized time-frequency masking in this paper. Although time-frequency masking is able to separate target speech from interferences effectively, one should consider two problems. One is that masking does not work well in noisy or reverberant environment. Another is that masking itself might cause some distortion of the target speech. For the former, we apply our time-frequency masking method [7] which can separate the target signal robustly even in noisy and reverberant environment. Next, investigating the distortion caused by time-frequency masking, we reveal following facts through experiments: 1) soft mask is better than binary mask in terms of recognition performance and 2) cepstral mean normalization (CMN) reduces the distortion, especially for that caused by soft mask. At the end, we evaluate the recognition performance of our method in noisy and reverberant real environment.",
keywords = "Blind source separation, Robust ASR, Speech sparseness, Stereo-input, Time-frequency mask",
author = "Yosuke Izumi and Kenta Nishiki and Shinji Watanabe and Takuya Nishimoto and Nobutaka Ono and Shigeki Sagayama",
year = "2009",
language = "English",
pages = "1955--1958",
journal = "Nuclear Physics A",
issn = "0375-9474",
publisher = "Elsevier",

}

TY - JOUR

T1 - Stereo-input speech recognition using sparseness-based time-frequency masking in a reverberant environment

AU - Izumi, Yosuke

AU - Nishiki, Kenta

AU - Watanabe, Shinji

AU - Nishimoto, Takuya

AU - Ono, Nobutaka

AU - Sagayama, Shigeki

PY - 2009

Y1 - 2009

N2 - We present noise robust automatic speech recognition (ASR) using sparseness-based underdetermined blind source separation (BSS) technique. As a representative underdetermined BSS method, we utilized time-frequency masking in this paper. Although time-frequency masking is able to separate target speech from interferences effectively, one should consider two problems. One is that masking does not work well in noisy or reverberant environment. Another is that masking itself might cause some distortion of the target speech. For the former, we apply our time-frequency masking method [7] which can separate the target signal robustly even in noisy and reverberant environment. Next, investigating the distortion caused by time-frequency masking, we reveal following facts through experiments: 1) soft mask is better than binary mask in terms of recognition performance and 2) cepstral mean normalization (CMN) reduces the distortion, especially for that caused by soft mask. At the end, we evaluate the recognition performance of our method in noisy and reverberant real environment.

AB - We present noise robust automatic speech recognition (ASR) using sparseness-based underdetermined blind source separation (BSS) technique. As a representative underdetermined BSS method, we utilized time-frequency masking in this paper. Although time-frequency masking is able to separate target speech from interferences effectively, one should consider two problems. One is that masking does not work well in noisy or reverberant environment. Another is that masking itself might cause some distortion of the target speech. For the former, we apply our time-frequency masking method [7] which can separate the target signal robustly even in noisy and reverberant environment. Next, investigating the distortion caused by time-frequency masking, we reveal following facts through experiments: 1) soft mask is better than binary mask in terms of recognition performance and 2) cepstral mean normalization (CMN) reduces the distortion, especially for that caused by soft mask. At the end, we evaluate the recognition performance of our method in noisy and reverberant real environment.

KW - Blind source separation

KW - Robust ASR

KW - Speech sparseness

KW - Stereo-input

KW - Time-frequency mask

UR - http://www.scopus.com/inward/record.url?scp=70450163599&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70450163599&partnerID=8YFLogxK

M3 - Article

SP - 1955

EP - 1958

JO - Nuclear Physics A

JF - Nuclear Physics A

SN - 0375-9474

ER -