The following information was submitted:
Transactions: WSEAS TRANSACTIONS ON INFORMATION SCIENCE AND APPLICATIONS
Transactions ID Number: 32-182
Full Name: Kun-Ching Wang
Position: Assistant Professor
Age: ON
Sex: Male
Address: No. 200, University Rd, Neimen Shiang, Kaohsiung, Taiwan, 845, R.O.C
Country: TAIWAN
Tel:
Tel prefix:
Fax:
E-mail address: kunching@mail.kh.usc.edu.tw
Other E-mails: wkc0224@seed.net.tw
Title of the Paper: Speech Detection Based on Entropy Measure for Variable Noise-Level in Cars
Authors as they appear in the Paper:
Email addresses of all the authors:
Number of paper pages: 10
Abstract: The goal of this study focuses on the presence of a robust method for speech detection (SPD). A robust SPD is essential part of speech recognition in non-stationary backgrounds such as in cars. It is shown that the inherent nature of banded structures appears only on speech-activity spectrogram and is robust against various types of noises. Consequently, a robust acoustic feature, namely band-splitting spectral entropy (BSE) parameter, is then proposed here. By use of the band decomposition, the banded structures are explicated more efficiently. To make the BSE parameter insensitive to colored noise such as car noise, the weighted average based noise estimator being used for tracking noise variation where the BSE decision is as an indicator of updating noise power is utilized. Experimental results show that the proposed SPD is robust against variable noise-level in cars.
Keywords: speech detection, spectral entropy, band-splitting
EXTENSION of the file: .pdf
Special (Invited) Session:
Organizer of the Session:
How Did you learn about congress:
IP ADDRESS: 123.204.152.235
