Research On The Speech Synthesis Of Tibetan Ando Dialect Based On HMM

With the development of artificial intelligence,research on computer speech synthesis has become widespread.However,speech synthesis techniques that allow computers to speak are more commonly used in Chinese,English,and Japanese.For some ancient ethnic minorities,speech synthesis technology still lags behind.Tibetan studies of Anduo dialects are based on linguistics.The study of speech synthesis technology in Tibetan dialects is still relativeiy scarce,and there is no set of computer tagging systems specifically for Ando dialects.This thesis mainly implements HMM-based speech synthesis of Tibetan dialects.The work and innovation of the thesis are as follows:Firstly,A set of computer annotation system for Tibetan speech synthesis in Anduo dialect was designed to achieve speech synthesis in Anduo dialect.Firstly,the most basic elementary vowels were used as research objects to sum up the characteristics of all the syllables in Anduo dialects.Then we studied the Tibetan language and designed the corresponding syllable and vowel bases based on the dichotomy.and designed the machine-readable phonetic transcription SAMPA-ADT of Anduo dialect.Combining word boundaries,sentence boundaries and other information in Anduo dialects.Completed the establishment of the tagging system.Secondly,Design a corpus of Tibetan dialects.According to the special pronunciation law of Anduo dialect,combined with the dialect of the dialect of Anduo dialect,500 text sentences of Ando were designed to find Ando 2 men and 2 women for professional recording.The linguistic data was divided and marked.Finally,Using the speech synthesis technology based on HMM,through speaker self-adaptive training,synthesized the Tibetan dialect of the dialect of high natural degree.Through speech synthesis experiments,MOS and DMOS evaluations were performed on synthesized speech.Experimental results show that speaker's adaptive speech training method can improve the sound quality of Tibetan dialect speech.