The Uygur Natural Spoken In Multiple Pronunciation Word Modeling And Performance Analysis

Speech recognition speech data into text sequence, is the main component ofhuman-computer speech interaction. The speech recognition technology from theinitial isolated digit recognition has developed to the stage of speaker independent,large vocabulary continuous speech recognition. The speech recognition technology,the research model and the language model and acoustic has been relatively mature,but the pronunciation model is still at the primary stage. Pronunciation model toimprove automatic speech recognition system recognition rate have a significantimpact, so the pronunciation model is a very important task. So this paper choosepronunciation modeling method as the research objective, and specifically on thepronunciation application model in Uighur natural accent continuous speechrecognition system has made the exploration and experiment. The main work of thispaper are as follows:1) the pronunciation variation occurs naturally in Uyghur language spoken, andsummed up the factors causing pronunciation variation, and built a small scale naturalaccent corpus, and in accordance with the speech corpus annotation specification forlabeling.2) studied the random Uighur natural accent prediction model based on. Thispaper established the Uyghur13000vocabulary pronunciation dictionary, thedictionary contains the standard pronunciation of each word and variation inpronunciation, and basic feature template is designed in this paper is established onthe conditional random field based on the Uyghur pronunciation prediction model. Atthe end of the Uyghur pronunciation prediction model on the test set predictionexperiment pronunciation.3) the Uyghur pronunciation performance prediction model in the Uygur naturalaccent continuous speech recognition system. This paper uses the Uygur language pronunciation prediction of pronunciation modeling of Uighur’s60000word model,and recognition experiments in spontaneous speech database, we can conclude fromthe experiment result, modeling method of Uyghur pronunciation model proposed inthis paper will be spontaneous speech recognition rate.