GMM Voice Conversion System Based Time Length Changed

The speech signal is only contains speech content information, but also includes speaker audio feature form information. Under the premise of guaranteeing pronunciation voice content does not change, through changing the personal characteristics of the source speaker’s voice, so that target speaker voice with personal characteristics,this technology called Voice Conversion.This technology encompasses almost every aspect of the speech signal processing field and it is one of the more popular technology currently. Prosodic features of voice has a great influence on naturally and intelligibility of the ultimate synthesized speech from conversion system.This paper proposed the algorithms of changes the time length of converted prosody features based on the traditional Gaussian mixture model speech conversion system, to make up the disadvantages of converted voice has low naturally compare with the post GMM model speech conversion system, it has improved the intelligibility of converted speech.In this thesis, we study some important technologies-based on GMM voice conversion system used, and then form objective evaluation and subjective evaluation to evaluate the converted speech,in order to determine how well the design of the conversion system is.The main work is as follows:1.The speech conversion system based on length transform is not only complete the basic voice conversion requirements, but also solve the shortcoming that the quality of synthetic voice through the system conversion is unnatural and rough and so on.From the vocal mechanism of voice began to study voice analysis model that suitable for voice conversion system, its corresponding speech parameters and conversion algorithm that used in voice conversion system. Focuses on the main voice conversion algorithm based on Gaussian mixture model voice conversion system and achieve simulation,at last,we get the subjective and objective test results.2. To the problem of low voice natural for traditional voice conversion system prevalent, we propose and implement an improved algorithm based on length changes voice conversion system, through change the time-scale of speech, which is operated with insert the converted parameters before and after each word. The results of the listening tests in which the naturalness and understandability of the converted voice are reported better than ever.3.In the voice conversion system based on the improved algorithm proposed before, MFCC is adopted to be extracted because it is more beneficial for sound perception. It is given the 3-D MFCC diagrams as well as waveforms of the voices before and after the conversion. The test results confirm that the transformed speech is not only approximates the characteristics of the target speaker, but also more natural and understandable.