Font Size: a A A

The Research On Vocal Tract Spectrum And Transition Methods In Voice Conversion

Posted on:2018-10-05Degree:MasterType:Thesis
Country:ChinaCandidate:T H DongFull Text:PDF
GTID:2348330536979566Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
The definition of Voice Conversion technology is that transformed the source speaker's voice characteristics to the goal speaker's voice characteristics without changing the content of speech.Speech conversion technique which belongs to a research direction of the Speech Signal Processing,it had the closer relations of speech signal analysis,identification,synthesis and promote the development of each other.In addition,there are many applications such as Text-to-speech conversion(TTS),Dubbing the film and television works,medical domain and so on.The details are as follows:Firstly,the role of the various parts in Speech Conversion System is discussed.Focused on the transformation of vocal tract spectrum which is one of voice characteristics and the key research content,introduced many of the transformation models.Secondly,because of Radial Basis unction(RBF)neural network has been used for converting vocal tract spectrum,its kernel function is being practiced with K-means Clustering.Due to slow convergence rate and easily trap in local optimum,a novel optimization method is proposed.An improved Particle Swarm Optimization(PSO)based method is proposed to optimize the centers of the RBF to improve the property of RBF network,hence enhance the transformation of speech parameters.The result,acquired by modeling and simulation,showed that the proposed method can effectively improved the performance of neural network and the converted voices are closer to the goal.Thirdly,Vocal Tract Spectrum of speech conversion system made use of single transform rule,whereas this method difficultly matched all the characteristic parameters which created a drop in the quality of the converted speech.In order to improve the situation,a measure of switching vocal tract spectrum based on Self-organizing Feature Map(SOFM)combined RBF neural network is presented.SOFM had good ability of classification to set up many kinds of transformation regulations.Subjective and objective experiments were carried out with the purpose of provoking the effect of the above methods.Draw a conclusion,This various kinds of mapping rule can improve the accuracy of the conversion,making the voice signal quality improved.
Keywords/Search Tags:Voice Conversion, Speech Signal Processing, Vocal Tract Spectrum Conversion, K-means Clustering, Radial Basis Function Neural Network, An Improved Particle Swarm Optimization, Self-organizing Feature Map
PDF Full Text Request
Related items