Digital Speech Rhythm Analysis Based On Segments Of Speech

With the rapid development of Internet technology, people are fully aware of the convenience of speech communication. There are many speech products on the market, for example, some speech synthesis software synthesis high quality speech. There is a need to adjust speech rhythm of some segments in continuous speech in many cases to achieve some special speech effect. Research on rhythm processing of speech segment are divide speech segments automatically and more accurate speech effect adjusting.The study is based on the speech segmentation, and thesis takes a thorough analysis on the algorithm, Chinese continuous speech endpoint detection method based on the vowel detection. The algorithm cost a lot of time in dividing the voice segment and the noise segment and many syllables are not divided in voice segment with this algorithm. Thesis proposes an improve algorithm for the two shortcomings of the algorithm.For adjusting the pitch characteristic of rhythm, thesis takes a thorough analysis on the autocorrelation function based pitch detection algorithm and the circulation average magnitude difference function based pitch detection algorithm. The two algorithms often get half or double frequency at pitch positions, which are caused by the influence of the interference waveform and the peaks of the waveform at the pitch positions are too weak to be covered by others. Thesis introduces a new algorithm to detect the pitch to improve the two shortcomings, to get more accurate pitch.Thesis mainly focuses on the following specific points:1. Analysis the difference between voice segments and noise segments in the time domain, and divide voice segments and noise segments by a small amount of calculation method, which is combine the energy-zero-ratio with thesis proposed adaptive threshold algorithm.2. In order to getting the boundary of each syllable, combine the number of syllables in voice segment with the idea of dichotomy to set a dynamically threshold, to get the same amount segments in voice segment with the number of syllables and then divide boundaries according to the characteristics of the transition energy between syllables.3. In order to reduce the interference waveform and increase the peak of the waveform at the pitch position, get a sharper waveform, using the ration of autocorrelation function and average magnitude difference function to detect the pitch of the speech.Experiments show that the algorithm proposed for syllable segment get a more accuracy result, and the mixed algorithm for pitch detection reduce the half frequency, double frequency error effectively.