Comparative Study Of Recognition Tools As Back-Ends For Bangla Phoneme Recognition
Authors
Arifa Ferdousi
Abstract
This paper deals with the development of a speech recognition system and comparative study of recognition
results for Bangla phonemes. At first, Phonemes were recorded and converted into digital form. Then MFCC
features from phonemes were extracted by Mel scale cepstral analysis. The recognition tools include Hamming
and Euclidean distance measurement and learning through a neural network. Ten Bangla phonemes were used
to test the system. The performance of the system shows that Euclidean distance measurement is the simplest
and better method in recognizing Bangla phonemes.
