The Design And Realization Of Mogolian Speech Synthesis System

With the continuous improvement of science and technology, people tend to find a new Human-computer interaction method like Speech, which was much more conform to the way of human communication. These subjects include speech recognition technology which can recognize human speech, and speech synthesis technology which can convert text to speech. At present, the theoretical research of Mongolian speech synthesis is already mature, but it only stayed in the experimental stage, and there isn’t a usable Speech synthesis system, especially in the part of front-processing, almost no other processing except the phoneme conversion. In order to improve the Mongolian People’s Human-computer interactive experience, in this paper we designed and implemented a Mongolian speech synthesis system with high Intelligibility and high naturalness score.According to the characteristics of the Mongolian pronunciation, we labeled 4709 sentence of Mongolian speech, using the recently famous speech synthesis model training tool —HTS (HMM-based speech synthesis system) trained a new Mongolian speech synthesis model, then realized the whole process of Mongolian speech synthesis front-processing including the special character processing, phoneme conversion, prosodic prediction, syllable division and label documents generation, etc. We build the web service of our Mongolian speech synthesis, which can provide service for any other system.In addition, we evaluated the Intelligibility and naturalness of our Mongolian speech synthesis model, we use SUS (Semantically Unpredictable Sentences) to test the Intelligibility, the correct rate for word dictation is 54.6%, and the MOS (Mean Opinion Score) of Subjective naturalness test reached 3.42. Therefore, the Mongolian speech synthesis system we build is reached a higher level in Intelligibility and naturalness.