Speech recognition using probabilistic and statistical models

This paper presents an implementation of probabilistic and statistical models for speech recognition. Three models namely Gaussian mixture model, hidden markov model and Gaussian mixture model - universal background model are discussed. In GMM, both speech identification of unknown isolated words and classification of unknown test patterns are discussed. In HMM, speech identification of isolated words are discussed. In GMM-UBM, speech identification of isolated words and speech classification of unknown test patterns are discussed. Isolated word recognizer build using all the three models for the recognition of isolated words can give 100% accuracy depending upon the initialization of the models. GMM-UBM is not found suitable for the classification of unknown test patterns.