Research On Speech Conversion From Tibetan Amdo Dialect To Tibetan U-tsang Dialect

With the rapid development of the Human-Machine speech interaction in recent years,the speech conversion technology is an important part of Human-Machine speech interaction.At present,speech conversion has been mainly concentrated in Mandarin in the domestic and has already implemented the speech conversion from Mandarin to Liaocheng dialect,Shenyang dialect,Xi'an dialect,Lanzhou dialect,etc.But it lacks of the speech conversion among various dialects of minority languages.Tibetan is one of the languages which is used for more people in China,so research on speech conversion among Tibetan dialects has a significant meaning.In this thesis,Tibetan Amdo dialect and U-tsang dialect are regarded as the research object for the study of the speech conversion among Tibetan dialects,a parallel voice corpus of Tibetan Amdo dialect and Tibetan U-tsang dialect are established.The thesis carries out an instrumental phonetics analysis of Tibetan Amdo dialect and Tibetan U-tsang dialect.Based on the instrumental phonetics analysis,the speech conversion model from Tibetan Amdo dialect to Tibetan U-tsang dialect is established.Eventually,the thesis uses the speech conversion model and STRAIGHT algorithm completed the research on speech conversion from Tibetan Amdo dialect to Tibetan U-tsang dialect.The main focus and originalities of the thesis are as follows:Firstly,a Tibetan corpus is established.A Tibetan text corpus is firstly deigned,and the thesis uses the Tibetan text corpus to record parallel speech corpus of Tibetan Amdo dialect and Tibetan U-tsang dialect,and then the thesis separates the segmentation,taggeds speech corpus and creats a directory index speech corpora of Tibetan Amdo dialect and Tibetan U-tsang dialect.It provides the most basic research data for the final completion of the speech conversion from Tibetan Amdo dialect to Tibetan U-tsang dialect.Secondly,the speech characteristics of Tibetan Amdo dialect and Tibetan U-tsang dialect are analyzed.In other words,the pitch analysis and pause duration analysis are accomplished for monosyllables and disyllables in Tibetan Amdo dialect and Tibetan U-tsang dialect.Than the thesis compares the analysis results,finds out the pitch and pause duration difference between Tibetan Amdo dialect and Tibetan U-tsang dialect.Thirdly,the monosyllables and disyllables speech conversion model from Tibetan Amdo dialect to Tibetan U-tsang dialect are established by the five-degree tone model,and the thesis completes the monosyllables and disyllables speech conversion from Tibetan Amdo dialect to Tibetan U-tsang dialect.Fourthly,the thesis uses the monosyllables and disyllables speech conversion model draw an initial pitch curve of the sentences speech conversion from Tibetan Amdo dialect to Tibetan U-tsang dialect what can be preliminarily generated.And the thesis uses the pitch compensation model and STRAIGHT algorithm further modified the initial pitch curve of the sentences.Eventually the thesis achieves the speech conversion from Tibetan Amdo dialect to Tibetan U-tsang dialect.