For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
Computational Complexity Reduction with Mel-Frequency Filterbank-Based Approach for Multichannel Speech Enhancement
Jungpyo HONG Sangbae JEONG
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences
Publication Date: 2017/10/01
Online ISSN: 1745-1337
Type of Manuscript: LETTER
Category: Speech and Hearing
complexity reduction, mel-filterbank, PMWF,
Full Text: PDF>>
Multichannel speech enhancement systems (MSES') have been widely utilized for diverse types of speech interface applications. A state-of-the-art MSES primarily utilizes multichannel minima-controlled recursive averaging for noise estimations and a parameterized multichannel Wiener filter for noise reduction. Many MSES' are implemented in the frequency domain, but they are computationally burdensome due to the numerous complex matrix operations involved. In this paper, a novel MSES intended to reduce the computational complexity with improved performance is proposed. The proposed system is implemented in the mel-filterbank domain using a frequency-averaging technique. Through a performance evaluation, it is verified that the proposed mel-filterbank MSES achieves improvements in the perceptual speech quality with a reduced level of computation compared to a conventional MSES.