Isolated Words Digits Speech Recognition

Abstract

This paper implemented a speech recognition program for isolated digit words using a method called the Hidden Markov Model (HMM) for speech modeling. The K-means,Baun-welch algorithms for training and codebook conception and finally the Viterbi decoding algorithm for recognition process. This method uses a statistical approach in characterizing speech. Briefly, speech utterance is fit into a probabilistic framework, which consists of transition of states and observable sequences. The target is to evaluate the probability score of the speech utterance based on a given model, and also to find the best model that gives the highest probability score. Research has shown that the HMM method is superior over conventional template matching methods, and it has already been applied by oversea companies successfully in commercial speech recognition programs. Implementing a LP Cepstrum, Coefficient function, a training function, which creates Hidden Markov Models of specific utterances and a testing function, testing utterances on the models created by the training-function. These functions created in MatLab. The recognized word decision is based on the maximal likehood value. The speech database is TI46 which is downloading from internet.