Hybrid HMM/ANN based isolated Hindi word recognition
Yogi Kapse, Narendra Digambar Londhe · 2014
Automatic speech recognition has become one of the most challenging task in the field of pattern recognition and natural language processing. In this paper, a hybrid model is proposed for isolated Hindi word recognition. This hybrid model involves the iterative training procedure. HMM is employed to induce the state transition probability distribution and ANN is employed as a classifier. HMM is designed by 4-state left to right model. In the proposed model ten Hindi words are used for the samples and five speakers for training and five distinct speakers for testing purpose and therefore the performance has achieved upto 89.8%.