The Study Of Speaker Voice Conversion Technology

The purpose of speaker voice conversion is to convert source of the speaker’s voice into thetarget speaker’s voice by changing the feature of the speech without changing content of it.Speaker voice conversion technology, as a new research direction of speech signal processingafter speech recognition, speaker recognition and speech synthesis technique, has importantapplication value. In this article, researches are mainly on the implementation of voiceconversion algorithm based on RBF artificial neural network and the improved Gauss modelin the following aspects of the work:In the course of the study, the feature parameters are emphasised on the spectral envelope,conversion of the spectral envelope, preprocessing, time domain analysis and linear predictionanalysis to the selected speech signal, using several signal processing methods for speechsignal feature parameters extraction, the analysis of the experimental results programming andsimulation on MATLAB. Because different of the characteristic parameter vector sequencesparameters, we using dynamic time warping algorithm for feature parameter vector sequencesaligned, and then using the RBF neural network and improved Gauss model to complete theconversion of voice signal, the subtraction clustering algorithm is referenced into the RBFneural network to study hidden layer, output layer are using particle swarm optimization algorithm for learning. linear transformation based on the posterior probability weighted in theGauss model and smoothing. Then, we use the STRAIGHT model for speech synthesis andmake contrast. The experimental results show that the conversions based on these twoalgorithms are effetely.