Isolated Myanmar Speech Recognition via ANN

Automatic Speech Recognition (ASR) is a popularand challenging area of research in human computerinteraction. This paper presents an isolated Myanmarspeech recognition system that is speaker dependentas well as speaker independent and developed byusing Artificial Neural Network techniques. In thissystem, the Mel Frequency Cepstral Coefficientsextracted from the manually preprocessed words areconsidered as the features to acoustically identify thespeeches. Those features are then used to train andtest the Backpropagation neural network model. Thissystem uses a database of 2800 utterances (names ofthe cities in Myanmar) by 10 talkers (4 males and 6females), from which 2400 utterances are used fortraining and 400 are used for testing and recognition.As per the experimental results, the proposed systemachieved the recognition rate of about 93.5% forknown speakers (i.e., speaker dependent) and 76.5%for unknown speakers (i.e., speaker independent)