| Summary: | 博士 === 國立高雄第一科技大學 === 工程科技研究所 === 100 === In this paper, we study the improvement of speaker verification, Chinese tone recognition and MP3 (MPEG-1/Audio Layer3) coding. In Chapter one, we introduce the characteristics of the linear prediction residual (LP residual), Mandarin tones and MP3 encoding principle. In Chapter two, we evaluate the usefulness of the LP residual for speaker recognition. In addition, we prove that LP residual is more useful than the linear predictive coding (LPC) in speaker verification, and may be complementary to LPC. In Chapter three, our study find that the pitch contour of tone 1 has slight up and down waves, while the pitch contours of the other three tones have very clear up and down waves. So we use the characteristics of tone 1 to propose a new method of Mandarin four tone recognition. In Chapter four, we investigate whether there is a transient signal in the time domain during MP3 coding process, and propose an algorithm of reducing the probability of making a wrong decision of switching to short blocks, to improve audio quality after MP3 coding. In Chapter five, we present the contribution of this paper and work in the future.
|
|---|
