Single channel speech enhancement

• Main Authors: Chen, Guo., Koh, Soo Ngee., Soon, Inn Yann.
• Other Authors: School of Electrical and Electronic Engineering
• Format: Research Report
• Published: 2008
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems
• Online Access: http://hdl.handle.net/10356/2845
• Institution: Nanyang Technological University

In this report, we present our research work on single channel speech enhancement. First, current major speech enhancement algorithms are reviewed and we conclude that the S-MMSE-LSAE algorithm is the best in overall performance among the well known algorithms reported in the existing literature and examined by the authors under most signal-to-noise ratios conditions. Secondly, an enhanced Itakura speech distortion measure is examined in our study. The proposed measure incorporates masking properties of the human auditory system into the original Itakura measure and substantially improves the correlation degree of objective measure with subjective evaluation. Next, an F-norm constrained SVD enhancement algorithm based on the conventional SVD-based algorithms is proposed. The traditional SVD algorithms are usually limited by the use of the fixed order of retained singular values, which are difficult to match diverse corrupted speech signals embedded in various noise environments.