Development Of Dai Language Text-to-speech Conversion System

With the development of mobile Internet,cloud computing and big data,more and more technology products are equipped with artificial intelligence.voice synthesis has been promoted as an important part of human-computer interaction.The process of speech synthesis is the process of converting information from text to speech.This contains a variety of calculations from the input text to the voice signal.To meet these computing needs,TTS system must have abstract linguistic analysis,voice coding and many other complex functions.Up to now,speech synthesis is mainly for Chinese,English and other common language,and in the field of minoriti’es language research is relatively lacking.In this paper,the Xishuangbanna Dai as a research language,design and implementation of the applicable Dai TTS application system.The main works in this thesis are as follows:1.This paper studies the text analysis and processing methods of Dai,and realizes the front-end text analysis and processing in Dai language translation system from the perspective of development.Text analysis and processing mainly include text normalization,word segmentation and text romanization.2.Construct a trained Dai acoustic model training system.This paper introduces the basic theory of speech synthesis based on hidden Markov model,and then studies the training method of acoustic model in Dai.In the process of model training,the extraction and proofing of acoustic parameters are introduced emphatically.Finally,on the Cygwin platform,HTS-2.0 is used to complete the training of the whole acoustic model,and the experiment of speech synthesis is based on the STRAIGHT synthesizer.3.Develop the Dai Text-to-Speech system.According to the preliminary research results,based on the Spring Web MVC framework,we can develop a Web application that can implement in-context text to speech.Front-end text analysis and processing functions and voice synthesis functions are integrated into the application.In this process,we used different calling methods for different modules to complete the Web application.The final experimental results show that the text analysis and processing module can basically meet the normalization and Romanization requirements of the Dai TTS system.The trained acoustic model can also synthesize the Dai speech that intelligibility and naturalness are acceptable.