Research On Articulatory Feature-based Speech Recogniton

Research show that the English speech recognition system based on articulatoryfeature（sAF）, such us place and manner of articulation, have been shown to be usefulunder varying circumstances.Then, the research of the AF_based Chinese Putonghuasystems became a hot point.In the xinjiang Uighur language speech recognitionsystems of using articulatory features research work, however, is still not a man to do.So, in this paper we first studied the AF_based Chinese Putonghua systems, andthen the development of a Uighur language speech recognition system usingarticulatory tandem features is presented.A set of six Articulatory Features(AF) thatare used for classifyng vowels and consonants of Uighur language is given, and theposteriors of these six AF classifiers are used as features in the HMM, compared withbaseline acoustic model using standard acoustic feature(MFCC).Experimental resultsshow that the speech recognition systems based on articulatory features of therecognition rate was significantly higher than the traditional MFCC features.The main works of this paper are as follows.◆Establish the mapping information of acoustic data to phones, and masterbasic mapping rules.◆Give a speech label, and mark each frame voice of the correspondingarticulatory features characteristics.◆The study of mandarin Chinese and uighur speech recognition system,establishing precise phonemes classifier.◆Use QuickNet tools for neural network(MLP) training, get the speech signalbelongs to all kinds of articulatory features of tandem characteristics.◆Using a baseline acoustic model, and this paper using standard39d MFCCas features in the HMM.◆Using a AF_based acoustic model, and testing the performance of the AF recognizer.◆Compared with baseline acoustic model using standard acoustic feature, anconclusion is given.