Speaker Identification Using Power Spectral Subtraction Method

Speaker Identification is the system which allows the user to access the system by making an utterance from the microphone and identifies the current talker out of a set of speakers. In daily life environment, noise is unavoidable and the speech signal recorded from the microphone is corrupted by the background noise. If the input speech signal is contaminated by noise, it happens that the performance and accuracy of the speaker identification system decrease. To solve this problem, power spectral subtraction method is used to eliminate the noise from the noisy input speech signal. In this paper, we emphasized the text-dependent speaker identification system. There are three main modules in the proposed system: background noise suppression, feature extraction and feature matching. Power spectral subtraction with Short-Time FourierTransform (STFT) is used for noise suppression. Mel-frequency Cepstral Coefficients (MFCC) is applied for feature extraction to extract the features from the enhanced speech signal that can later be used to represent each speaker. For feature matching, LBG Vector Quantization (VQ) approach is proposed because it can reduce the amount of data and complexity. The experimental results show that the proposed system is more accurate than the original system and faster in computation than the original one. MATLAB is used for programming to simulate the proposed system.