High-performance speech recognition for TVs with NLMS and beamforming

Author

Hong, Jungpyo ; Park, Sangjun ; Lim, Jong-Jin ; Lee, Bu-Yeol ; Hahn, Minsoo

Author_Institution

Dept. of Electr. Eng., Korea Adv. Inst. Sci. & Technol., South Korea

fYear

2012

fDate

13-16 Jan. 2012

Firstpage

263

Lastpage

264

Abstract

In order to remove TV noise effectively, a novel system, combination of single channel and multi-channel microphone-based noise reduction method, is proposed. A single channel filtering based on normalized least mean square is performed on each channel to reduce TV noise using a reference noise provided by TV speaker output. After that, as a post-processing, generalized sidelobe canceller attenuates more the remaining nonstationary noises in input speeches. For performance evaluation, a HMM-based speaker recognition rate is measured and the recognition rate showed remarkable improvement.

Keywords

filtering theory; hidden Markov models; least mean squares methods; signal denoising; speaker recognition; HMM-based speaker recognition rate; TV noise removal; TV speaker output; generalized sidelobe canceller; high-performance speech recognition; input speeches; multichannel microphone-based noise reduction method; nonstationary noises; normalized least mean square; performance evaluation; post-processing; reference noise; single channel filtering; Delay effects; Microwave integrated circuits; Noise; Noise reduction; Speech; Speech recognition; TV;

fLanguage

English

Publisher

ieee

Conference_Titel

Consumer Electronics (ICCE), 2012 IEEE International Conference on

Conference_Location

Las Vegas, NV

ISSN

2158-3994

Print_ISBN

978-1-4577-0230-3

Type

conf

DOI

10.1109/ICCE.2012.6161858

Filename

6161858