Abstract—In this paper, a novel speaker feature extraction with Gaussian Mixture Model (GMM) is proposed. In this new method Perceptual Linear Prediction (PLP) and Linear Predictive Cepstral Coefficient (LPCC) features have extracted and Gaussian Mixture Models (GMM) of speakers has built, then identification tests with clean and noisy TIMIT database have been carried out. With usage of TIMIT database, train and test samples of the speech ratio is 9 to 1. Implementation results with GMM have shown that GMM will model the structure of the vocal tract finely and minimize the distance between training and test feature vectors. Also experimental results show that LPCC feature coefficients will improve the results of speaker recognition rate. Thus in new proposed method with combination of PLP and LPCC features, the efficiency of the speaker recognition rate will increase 2.2% and speaker recognition efficiency will be 98.4%.

 

Index Terms—speaker recognition, Gaussian mixture model, feature extraction, expectation maximization, TIMIT database

Cite: Masood Qarachorloo and Gholamreza Farahani, "New Proposed Feature Extraction Method to Enhance Speaker Recognition Rate with GMM," International Journal of Signal Processing Systems, Vol. 4, No. 4, pp. 276-281, August 2016. doi: 10.18178/ijsps.4.4.276-281