Beamforming networks using spatial covariance features for far-field speech recognition

Xiong Xiao; Shinji Watanabe; Eng Siong Chng; Haizhou Li

doi:10.1109/APSIPA.2016.7820724

Beamforming networks using spatial covariance features for far-field speech recognition

Xiong Xiao, Shinji Watanabe, Eng Siong Chng, Haizhou Li

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

11 Citations (Scopus)

Abstract

Recently, a deep beamforming (BF) network was proposed to predict BF weights from phase-carrying features, such as generalized cross correlation (GCC). The BF network is trained jointly with the acoustic model to minimize automatic speech recognition (ASR) cost function. In this paper, we propose to replace GCC with features derived from input signals' spatial covariance matrices (SCM), which contain the phase information of individual frequency bands. Experimental results on the AMI meeting transcription task shows that the BF network using SCM features significantly reduces the word error rate to 44.1% from 47.9% obtained with the conventional ASR pipeline using delay-and-sum BF. Also compared with GCC features, we have observed small but steady gain by 0.6% absolutely. The use of SCM features also facilitate the implementation of more advanced BF methods within a deep learning framework, such as minimum variance distortionless response BF that requires the speech and noise SCM.

Original language	English
Title of host publication	2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9789881476821
DOIs	https://doi.org/10.1109/APSIPA.2016.7820724
Publication status	Published - 2017 Jan 17
Externally published	Yes
Event	2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016 - Jeju, Korea, Republic of Duration: 2016 Dec 13 → 2016 Dec 16

Other

Other	2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016
Country/Territory	Korea, Republic of
City	Jeju
Period	16/12/13 → 16/12/16

ASJC Scopus subject areas

Artificial Intelligence
Computer Science Applications
Information Systems
Signal Processing

Access to Document

10.1109/APSIPA.2016.7820724

Cite this

Beamforming networks using spatial covariance features for far-field speech recognition. / Xiao, Xiong; Watanabe, Shinji; Chng, Eng Siong et al.

2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016. Institute of Electrical and Electronics Engineers Inc., 2017. 7820724.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Xiao, X, Watanabe, S, Chng, ES & Li, H 2017, Beamforming networks using spatial covariance features for far-field speech recognition. in 2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016., 7820724, Institute of Electrical and Electronics Engineers Inc., 2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016, Jeju, Korea, Republic of, 16/12/13. https://doi.org/10.1109/APSIPA.2016.7820724

@inproceedings{9ba83e08b86f4a09b802cd865ba3867c,

title = "Beamforming networks using spatial covariance features for far-field speech recognition",

abstract = "Recently, a deep beamforming (BF) network was proposed to predict BF weights from phase-carrying features, such as generalized cross correlation (GCC). The BF network is trained jointly with the acoustic model to minimize automatic speech recognition (ASR) cost function. In this paper, we propose to replace GCC with features derived from input signals' spatial covariance matrices (SCM), which contain the phase information of individual frequency bands. Experimental results on the AMI meeting transcription task shows that the BF network using SCM features significantly reduces the word error rate to 44.1% from 47.9% obtained with the conventional ASR pipeline using delay-and-sum BF. Also compared with GCC features, we have observed small but steady gain by 0.6% absolutely. The use of SCM features also facilitate the implementation of more advanced BF methods within a deep learning framework, such as minimum variance distortionless response BF that requires the speech and noise SCM.",

author = "Xiong Xiao and Shinji Watanabe and Chng, {Eng Siong} and Haizhou Li",

year = "2017",

month = jan,

day = "17",

doi = "10.1109/APSIPA.2016.7820724",

language = "English",

booktitle = "2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

address = "United States",

note = "2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016 ; Conference date: 13-12-2016 Through 16-12-2016",

}

TY - GEN

T1 - Beamforming networks using spatial covariance features for far-field speech recognition

AU - Xiao, Xiong

AU - Watanabe, Shinji

AU - Chng, Eng Siong

AU - Li, Haizhou

PY - 2017/1/17

Y1 - 2017/1/17

N2 - Recently, a deep beamforming (BF) network was proposed to predict BF weights from phase-carrying features, such as generalized cross correlation (GCC). The BF network is trained jointly with the acoustic model to minimize automatic speech recognition (ASR) cost function. In this paper, we propose to replace GCC with features derived from input signals' spatial covariance matrices (SCM), which contain the phase information of individual frequency bands. Experimental results on the AMI meeting transcription task shows that the BF network using SCM features significantly reduces the word error rate to 44.1% from 47.9% obtained with the conventional ASR pipeline using delay-and-sum BF. Also compared with GCC features, we have observed small but steady gain by 0.6% absolutely. The use of SCM features also facilitate the implementation of more advanced BF methods within a deep learning framework, such as minimum variance distortionless response BF that requires the speech and noise SCM.

AB - Recently, a deep beamforming (BF) network was proposed to predict BF weights from phase-carrying features, such as generalized cross correlation (GCC). The BF network is trained jointly with the acoustic model to minimize automatic speech recognition (ASR) cost function. In this paper, we propose to replace GCC with features derived from input signals' spatial covariance matrices (SCM), which contain the phase information of individual frequency bands. Experimental results on the AMI meeting transcription task shows that the BF network using SCM features significantly reduces the word error rate to 44.1% from 47.9% obtained with the conventional ASR pipeline using delay-and-sum BF. Also compared with GCC features, we have observed small but steady gain by 0.6% absolutely. The use of SCM features also facilitate the implementation of more advanced BF methods within a deep learning framework, such as minimum variance distortionless response BF that requires the speech and noise SCM.

UR - http://www.scopus.com/inward/record.url?scp=85013826101&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85013826101&partnerID=8YFLogxK

U2 - 10.1109/APSIPA.2016.7820724

DO - 10.1109/APSIPA.2016.7820724

M3 - Conference contribution

AN - SCOPUS:85013826101

BT - 2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2016 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference, APSIPA 2016

Y2 - 13 December 2016 through 16 December 2016

ER -

Beamforming networks using spatial covariance features for far-field speech recognition

Abstract

Other

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this