Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/29474
Title: A hybrid design of beamformers for voice control devices
Authors: Yiu, KFC 
Chan, KY
Grbic, N
Nordholm, S
Keywords: Beamformer
Microphone array
Multi-channel filter
Noise reduction
Signal enhancement
Speech recognition
Issue Date: 2012
Publisher: Yokohama Publishers
Source: Pacific journal of optimization, 2012, v. 8, no. 3, p. 533-544 How to cite?
Journal: Pacific journal of optimization 
Abstract: In this paper, a new approach to designing beamformers for voice control device is proposed. It is well-known that under a strong near-field noise with low signal-to-noise ratios (SNR), the performance of speech recognition is deteriorated significantly. However, designing the beamformer for enhancing speech recognition is a slow process and might not be adapted easily to the changing noise environment. In order to lower the complexity of the design, we intend to exploit the combination of existing optimal beamformer designs, which can be implemented in parallel. These include the least-squares technique and the signal-to- noise ratio maximization technique. We show here that for a given pre-trained speech recognizer and for a finite set of speech commands, neither method has a satisfactory performance in speech recognition accuracy under very low SNRs. However, since the two techniques have different characteristics in speech distortion and noise suppression, we show that it is possible to enhance the speech recognition accuracy by combining these two optimal designs.
URI: http://hdl.handle.net/10397/29474
ISSN: 1348-9151
EISSN: 1349-8169
Appears in Collections:Journal/Magazine Article

Access
View full-text via PolyU eLinks SFX Query
Show full item record

SCOPUSTM   
Citations

1
Citations as of Jul 7, 2017

WEB OF SCIENCETM
Citations

1
Last Week
0
Last month
0
Citations as of Aug 15, 2017

Page view(s)

45
Last Week
3
Last month
Checked on Aug 13, 2017

Google ScholarTM

Check



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.