Researching Of The Mongolian Acoustic Model Based On Speech Recognition

Speech recognition is technology that machine transforms voice signal of human into corresponding text or order through recognition and understanding process. As an important research subject in artificial intelligence, its' development will bring enormous influence to future man-machine interaction. In speech recognition research, non-specific person,large vocabulary, continuous speech recognition is most difficult and challenging topic. Mongolian is an influential language in the world, research of Mongolian speech recognition will certainly promote the development of Mongolian language information processing, so it has vital practical significance.Which sentence is the most possible one under a given pronunciation? Usually, this problem may transform to calculate two probability product of each sentence .These two product items are the linguistic model and acoustic model. The linguistic model gives the probability of random word sequence which appears in the text; it can restrain the search space, enhance the cutting accuracy, and thus reduce the rate of misrecognition. The acoustic model describes probability of acoustic sequence under some kind of word sequence; it is not only the fundamental model in recognition system, but also the most essential part.In order to enhance the performance of acoustic model, one way is to design the pronunciation database script carefully and increase the pronunciation samples unceasingly, so the sound corpus will cover all possible acoustic and linguistic phenomenon as far as possible. Another way is to research better modeling technology to enhance the accuracy of model, such as classified modeling and environment related modeling; this article mainly studies this aspect.