Digital signal processing : enhancing speech signal

• Main Author: Tran, Xuan Anh
• Other Authors: Chng Eng Siong
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Software::Programming techniques
• Online Access: http://hdl.handle.net/10356/59995
• Institution: Nanyang Technological University

The objective of this report is to discuss about di erent speech enhancement algorithms and evaluate them. Speech enhancement is very important as it is needed in all speech processing systems to reduce background noise before speech arrives to the listener. There are many speech enhancement algorithms created in the past few decades, and each of them always has its own strength and weakness. Understanding these techniques is necessary to have better choices in di erence cases as well as to suggest future work for improvement. In this report, the content is split into 5 chapters. Several fundamental knowledge will be discussed in order to have better understanding about the report. After that, two commonly used approaches for speech enhancement will be analysed in detail. They are spectral subtraction and minimum-mean- square-error (MMSE) estimator. Each of them has its own advantages over the other. While spectral subtraction is very simple and easy to implement, MMSE estimator appears to be more e cient in reducing both residual and musical noise for high or even low signal-to-noise speeches. Furthermore, the implementations of these two techniques in Matlab are performed in order to have better and more reliable evaluation. In conclusion, the results obtained by this experiment agree well with what is given by the theory. Speech enhancement is a very challenging task in a past few decades. There is normally trade o when too much noise is suppressed from the noisy signal, because speech will be easily distorted, resulting in poor speech intelligibility. Researchers have still been working on it to have the optimal algorithm which improves not only speech quality but also speech intelligibility.