Dialectal Assamese vowel speech detection using acoustic phonetic features, KNN and RNN

The recognition of vowel phonemes plays an important role in the field of speech processing. Assamese is the major language of Assam and also the mother-tongue of the of the largest segment of the population of Assam. The standard Assamese language has four major dialects namely Central dialect, Eastern dialect, Goalpariya dialect and Kamrupi dialect. It has eight vowel phonemes which are /i/, /e/, /ε/, /a/, /n/, /c/, /o/ and /u/. In this paper, a comparative analysis between the Recurrent Neural Network (RNN) based algorithm and K-Nearest Neighbor (KNN) based algorithm is carried out for the recognition of the vowel sounds using the Acoustic Phonetic Features as the feature vector. Dialect wise recognition of the vowels is also carried out using the same feature vectors. A recognition rate of 97 % is obtained by using the KNN based algorithm for vowel recognition and an overall rate of 84.3% and 87% is obtained by RNN based algorithm and KNN based algorithm respectively for the dialectal Assamese vowel recognition. K-NN based approach gives better recognition rate than the ANN based approach.