Word And Speaker Recognition System

In this report, a system which combines user dependent Word Recognition and text dependent speaker recognition is described. Word recognition is the process of converting an audio signal, captured by a microphone, to a word. Speaker Identification is the ability to recognize a person identity base on the specific word he/she uttered. A person's voice contains various parameters that convey information such as gender, emotion, health, attitude and identity. Speaker recognition identifies who is the speaker based on the unique voiceprint from the speech data. Voice Activity Detection (VAD), Spectral Subtraction (SS), Mel-Frequency Cepstrum Coefficient (MFCC), Vector Quantization (VQ), Dynamic Time Warping (DTW) and k-Nearest Neighbour (k-NN) are methods used in word recognition part of the project to implement using MATLAB software. For Speaker Recognition part, Vector Quantization (VQ) is used. The recognition rate for word and speaker recognition system that was successfully implemented is 84.44% for word recognition while for speaker recognition is 54.44%.