Research On Speech Signal Processing Technology

As lately developed techniques in man-machine area,the two primary aspects of speech signal processing,Time Scale Modification of speech and Voice Modification are researched in this paper.Time Scale Modification of speech is on the premise of unchanging tone of voice and to ensure good voice quality,changing speech pace freely. It is as if to speak in their own words to change the speed of the words.Voice Modification is a relatively new branch of voice processing;mainly through the control and adjustment appropriate parameters of synthesis voice to achieve.This paper studies the Synchronous Overlap-and-Add Fixed Synthesis(SOLAFS) algorithm for Time Scale Modification of speech.The proper choices of primary parameters are given experimentally.The results show that the synthesis shift(S_s) should be neither shorter than the average pitch nor longer than the subtraction of the average pitch from window length.The proper value of shift search interval(k_max) is equal to that of the average pitch approximately.Based on SOLAFS algorithm,the human auditory perception to pitch synchronization is investigated,which shows that the error of perception exists.The algorithm based on power spectrum reprocessing is modified which improves the accuracy of pitch detection,and a novel TSM algorithm based on it is given,which provides superior speech quality and reduces the computation complexity.The acoustical differences of pitch,a novel pitch modification algorithm based on modified power spectrum reprocessing is proposed,which reduces the computation complexity.A non-proportioned modification method for the first three formants is given,preserving the distribution of speech energy.A compensation technique for spectral tilt is suggested which improves the naturalness of output speech in the conversion of male to female.Finally a voice modification system based on cepstrum analysis is realized by integrating all proposed methods,and a sort of practical speech signal processing software is developed finally.The software used programming ideas mixed with VC++ and Matlab.The software can achieved change of voice speed and tone for input voice.In the changing of tone,the voice can be rapidly changed between male,female, Children and elder.It can also be manually set parameters,and realize other requirements of other voice changes,such as the voice changing between the same sexes.