基于lpc的语音旌旗灯号剖析分化的参考资料(国外英文资料).docVIP

基于lpc的语音信号分析合成的参考资料 Based on LPC voice signal analysis synthesis The frame of the speech signal Speech signal is one of the most important features: a short, in some timely section presents the characteristics of random noise, the others present a periodic signal characteristics in short period, there is a mixture of both. The characteristics of the speech signal is changed over time, only in a short time interval of voice signal to maintain relatively stable and consistent characteristics. This paragraph of time often take 5 ~ 50 ms. About 256 sampling points in the program, for 8 k sampling frequency, the equivalent of 32 Ms. Overlap between frame as the 128 sampling points, 16 ms. 1.2 endpoint detection Chinese syllables are voiced at the end, only short-term energy can better judge a word at the end of the point. Of course, sometimes terminal drag too long, decays more slowly, sometimes terminal attenuation is faster, it is a bit error. General as long as the short time average magnitude value reduced to the syllables of short time average magnitude largest 1/16 or so later, you can think the syllable has ended. Actually cut off a bit the tail will not significantly affect the recognition and synthetic processing. So there is no Chinese isolated words at the end of the detection difficult. Once the starting point of Chinese speech detection not only have the difficulty, and detect whether the accurate influence on the performance of speech recognition, because most of the initials are clear, and air supply and unaspirated plosive and fricative, with their environmental noise resolution is more difficult. Heres a look at the relevant technologies for voice endpoint detection: 1.2.1 short term energy analysis The definition of short term energy is as follows: The short term energy of a window function at the NTH point of the signal. You can see that the short term energy can be seen as the output of a linear filter by the square of the speech signal. Principle of