Research And Implementation Of Voice Conversion Techniques Based On ARM9

Voice conversion is a technique for converting the characteristic of speaker, which modifies the speech signal of a source speaker so that it sounds as if it is uttered by a target speaker. As a relatively new branch of speech signal processing, the studies on it will promote the research of speech analysis, speech synthesis, speech coding, speech recognition, speech enhancement and so on. Because voice conversion has been widely applied to text-to-speech, secure communication, interpret telephony and so on, it has important applications both at theories and practices.In this thesis, two key technologies of voice conversion pitch-scale modification and spectral envelope conversion are investigated. A system of voice conversion has been designed based on ARM9. The main works are as follows:(1) Some methods of pitch-scale modification, for instance, PSOLA, interpolation and decimation, spectral compression and expansion and harmonic copy, are studied. Besides, the simulation results are shown in the thesis.(2) The pseudo harmonic analysis-synthesis model is discussed and analyzed in detail. The pitch-scale and time-scale modification is achieved based on the model. At last, the simulation results of voice conversion are analyzed.(3) The some methods for transforming the spectral envelope, such as resampling, neural net, and bilinear function are studied. In the thesis, the bilinear function is adopted.(4) The critical dumping filters are discussed, with which the spectral tilt compensation is achieved. The frequency response of filters is introduced. At last, the simulation results of voice conversion are given.(5) A system has been proposed to fulfill the voice conversion between men and women based on the LPC analysis, which have compared with other methods, In addition, the result of subjective test is given.(6) The voice conversion system has been implemented on the Samsung S3C2410 microprocessor.The system given in the thesis is easy to implement and operate. The simulation result shows that the system can convert the speaker individuality successfully with high quality.