Research And Implementation Of Speech Synthesis Algorithm Based On Improved MFCC

Speech synthesis technology,as a text-speech conversion system,generates understandable and naturally sound artificial speech technology for a given input text.The technology is widely used in the fields of voice dialogue system,AC robot,singing speech synthesizer and speech speech translation system.With the continuous development of scientific and technological technology,speech synthesis methods are increasing day by day.Although the speech synthesis based on deep neural network is also beginning to obtain preliminary results,there are traditional methods.In this paper,the MFCC coefficient and the model training and synthetic speech based on HMM speech synthesis algorithm are presented,and the problems such as low natural degree of speech in the final synthesis are obtained.In recent years,improving the natural degree of speech synthesis has become a research hotspot in the field of speech synthesis.For the natural degree of synthetic speech,the following is studied:(1)Using the speech synthesis method based on HMM,the feature extraction part of the speech corpus processing stage is improved,and the improved MFCC feature extraction method is proposed to improve the effect of speech synthesis algorithm.In this research problem,the main innovation point of this topic is to propose an adaptive shrinkage reconstruction algorithm based on Shrink wavelet transform by means of wavelet transform,which can obtain better speech information for different speech signals,thus improving the accuracy of MFCC feature samples.(2)In this paper,the speech library is extracted based on the above improved MFCC feature extraction,and the results are applied to the simulation of speech synthesis algorithm of HMM training.The experimental results show that the mean square root error(RMSE)index of the MFCC parameter of synthetic speech is compared with the mean square root error(RMSE)index of the MFCC parameter of the original speech signal,which has an average performance improvement of about 19%.Finally,subjective evaluation method was used to find that the natural degree of synthetic speech was increased by 10%.The research in this paper is mainly to further study and provide the scheme for the natural degree of speech synthesis.The results of the project can be applied in the field of speech signal processing.