MULTIPLE SPEAKERS SPEECH RECOGNITION FOR SPOKEN DIGITS USING MFCC AND LPC BASED ON EUCLIDEAN DISTANCE

Abstract

The fundamental part of human life that acts as one of the five senses of human body is speech recognition, for this reason the applications that come forth on the basis of speech recognition, have high level of favourable reception. This paper has endeavoured to examine multiple steps encompassed in artificial speech recognition by man-machine interface. In speech recognition multiple steps that we persued are distance calculation, feature extraction, dynamic time wrapping. Examining the similarity measuring algorithms in ASR systems was the most comprehensive aim of the proposed research. The feature vectors such as LPC, and MFCC were evaluated in the first place. Once the operations were carried we probed MFCC specifically. Moreover it was designated as the preferred mode of feature vector coding as they follow the human ear’s reaction to the sound signals. Many new techniques of distance measurement were established and a comparison was drawn between them and thus the conclusion was made that Euclidean distance measure is a favoured one when the template database of sound is very poor. We carried out a rapid investigation of dynamic time wrapping algorithm and found the slightest path between two sounds. Then we devised a compact model by writing a simple code which was fit to identify small set of isolated words. The proposed speech recognition methodology has been implemented for multiple speakers also.