Research And Implementation Of Sequence To Sequence Tibetan Lhasa Dialect Speech Synthesis

Nowadays,with the quick development of Information technology,Artificial Intelligence(AI)is in the ascendant.Speech processing technology has got distinguishing achievements stimulated by the development of new era.Speech synthesis technology has grown from the initial mechanical and electronic devices to utilizing parameter statistics,waveform splicing,deep learning and so on.Over the last two years,the end-to-end technology in speech synthesis has come into use,and it makes it possible to employ the same model from the step of training to synthesis.End-to-end technology is able to make inputs from one end and get outputs directly from the other end.Sequence-to-sequence technology,as one of the classical end-to-end speech synthesis methods,has made it come true to input texts and output waveform files directly.Based on the background of Tibetan Lhasa dialect speech synthesis technology,this paper introduces the development process of speech synthesis technology and Tibetan speech synthesis technology,as well as the theoretical researches carried out on Tibetan speech synthesis and the latest technical developments.Making use of application of sequence-to-sequence model structure to realize the Tibetan speech synthesis technology,this paper analyzes the current situations of Tibetan speech synthesis technology and puts forward expectations of further development and improvement.The synthesized speech by sequence-to-sequence model structure is almost indistinguishable from that of a real speaker.Compared with traditional HMM and DNN-HMM,it has improved a lot and achieved the MOS score of 3.92 and the similarity score is 4.20.In addition,Chinese experiments were conducted to analyze the aspects to be improved in Tibetan corpus.The research on sequence-to-sequence Tibetan Lhasa speech synthesis has the advantage of saving a lot of labor and time cost in the front end,which,to some extent,has stimulated the wide application of speech synthesis technology.Sequence-to-sequence Tibetan Lhasa dialect speech synthesis research has opened up a new research area for many computer technology researchers and provides a promising future of getting more inter-disciplinary achievements.