JAIST Repository >
School of Information Science >
Conference Papers >
Conference Papers >

Please use this identifier to cite or link to this item: https://hdl.handle.net/10119/9966

Title: An MTF-based Blind Restoration Method for Improving Intelligibility of Bone-conducted Speech
Authors: Kinugasa, Kota
Unoki, Masashi
Akagi, Masato
Issue Date: 2009-03-01
Publisher: 2009 International Workshop on Nonlinear Circuits and Signal Processing
Magazine name: 2009 International Workshop on Nonlinear Circuits and Signal Processing (NCSP'09)
Start page: 105
End page: 108
Abstract: Bone-conducted (BC) speech is useful for speech communications in extremely noisy environments. However, sound quality and intelligibility of BC speech are very poor and low. In this paper, we propose a blind restoration method for improving BC speech based on the concept of modulationtransfer function (MTF). We first investigate the relationship between the power envelope of air-conducted (AC) and BC speech signals by analyzing AC/BC database. We then model these relations by fitting some models to the MTF derived from the database. The best-fitted model has two parameters.One is the gain factor, and another is the attenuation factor. We then propose a method for determining the parameters of model without the AC speech. We finally evaluate the proposedmethod using SNR, correlation, RMS of MTF, PESQ,and LSD. As the results, we showed the proposed method is effective for blind BC speech restoration.
Rights: This material is posted here with permission of the Research Institute of Signal Processing Japan. Kota Kinugasa, Masashi Unoki, and Masato Akagi, 2009 International Workshop on Nonlinear Circuits and Signal Processing (NCSP'09), 2009, pp.105-108.
URI: https://hdl.handle.net/10119/9966
Material Type: publisher
Appears in Collections:b11-1. 会議発表論文・発表資料 (Conference Papers)

Files in This Item:

File Description SizeFormat
NCSP2009_Kinugasa.pdf255KbAdobe PDFView/Open

All items in DSpace are protected by copyright, with all rights reserved.

 


Contact : JAIST Library (ir-sys[at]ml.jaist.ac.jp)