Research On Random Speech Signal Synthesis Method

With the rapid development of science and technology,conference recording interference field had proposed new demands for random speech signal synthesis algorithm.Therefore,this paper launched a random speech signal synthesis algorithm research.At present,the synthesis of speech signals has a mature theory,but all are deterministic methods,that is,once the speech synthesis parameters are determined,the output is repeated multiple times and the output is unchanged.It cannot achieve requirements that multiple times synthetic output and the results random changes.About the random speech synthesis algorithm,there is not literature can be a direct reference,and this study corresponds to this new application.On the basis of the introduction of linear predictive speech synthesis algorithm and artificial neural network model by using speech library to synthesize random speech storage space,this propose a new method of random speech synthesis based on quadratic compression.Respectively,linear predictive speech synthesis algorithm and artificial neural network model for primary and secondary compression are used to reduce the storage needs of the voice library.At the same time,the voice library data is chosen,which leaves only voiced and removes the voiceless.And then it use the random starting point and random length to produce the required random speech,the experiment verifies the effectiveness of the method.Put forward the mos of random speech.Secondly,the realization of random speech synthesis algorithm based on quadratic compression is studied.Firstly,the algorithm use matlab to achieve and test.In order to transplant the algorithm to the embedded system,this paper study the code transplant method of C language,and it summed up a set of effective method of the process,the innovative solution to the practice of some of the code conversion optimization issues.Experimental verifies that it valid.Finally,the method of random speech synthesis based on formant is proposed to solve the shortcoming which the above algorithm rely on voice library and two improved algorithmalgorithms are are proposed.Firstly,a new algorithm of random speech signal synthesis based on formant is constructed by transforming the formant parameters to achieve random selection.Experimental verifies that it can produce random speech,but the output pass rate is not high,the most of the output sound are non-voice.Furthermore,an improved algorithm based on the ratio of the formant parameters is proposed,and the experimental show that the output pass rate of random speech is improved.Then a new algorithm of stochastic speech synthesis based on spectrogram decision and parametric ratio is proposed.The experiment proves the effectiveness of the algorithm.