| Summary: | 碩士 === 國立臺灣大學 === 工程科學及海洋工程學研究所 === 100 === The cell phone technology is rapidly making its changes, adding more and more variety to its functions. But the most essential demand from the cell phone users is merely an excellent quality of communication. However, it is rare for most of us to converse over the phone without a hint of background noise or the likes, forcing us, as a result, to have misunderstandings or a void in our communication. In this thesis, we will attempt to utilize an adaptive algorithm to optimize the information collected from the dual-microphone.
Four of the speech signals were selected from the MAT Speech Database, preserved by the Computational Linguistics and Chinese Language Processing Association, to serve as mock speeches for the speakers in this observation. Furthermore, a signal based on the features of the time domain and the frequency domain was utilized, along with the help of the end-point detector, to find out the amount of energy and entropy needed to determine the location of a signal''s starting and ending point. Then through this, the signal was able to be determined before it is received. As for the employment of the adaptive algorithm, this thesis has cited a number of periodicals and theses to help readers gain a better understanding of the features of the active noise cancellation system. Moreover, this thesis adopted the least mean square algorithm, the normalized least mean square algorithm, and the wavelet based least mean square algorithm as expedients in eliminating background noises. Because our environment is filled with noises of various frequencies or energy, and in order to recognize the pros and cons of the different algorithms being used to eliminate the broadband and narrowband interferences, the sinusoidal signal with the frequencies at 200Hz, 1000Hz, and 2000Hz was implemented to test out how well and efficient the algorithms are able to eliminate the narrowband noises. This test will further attempt to exclude the Gaussian white noise and the Gaussian pink noise in the broadband. The two methods, the least mean square algorithm and the normalized least mean square algorithm, both directly eliminated noise after the dual-microphone had collected the signals. However, it was observed that the adaptive algorithm served to better eliminate the narrowband noises. Therefore, the Wavelet based least mean square algorithm was utilized to decompose, de-noise, and recompose the signals in attempt to parallelize the calculation process. The software, MatLab, was also employed to stimulate the possibilities when the algorithm is applied to the actual hardware.
The result showed that the adaptive algorithm had great effects upon the elimination of both narrowband and broadband noises, and it efficiently preserved the completion of the original speech signals to minimize the distortion. When the signal was longer, the adaptive algorithm was observed to perform a greater convergence effect through number of iterations. Therefore, the algorithm had a greater performance on the function of de-noising.
|
|---|
