Abstract—In this paper, the new method to improve speaker recognition efficiency has been proposed. In this new method firstly, silence of the sentences has removed and secondly different train and test data with -5, 0, 5 and 10 dB signal to noise ratio has made artificially. Spectral Subband Centroid (SSC) and Linear Predictive Cepstral Coefficient (LPCC) features were extracted and then Gaussian Mixture Models (GMM) of speakers has built and identification tests with clean and noisy TIMIT database have been used. In the used TIMIT database, train and test samples of the speech at a ratio of 9 to 1 are used. Implementation results with comparison between different characteristics for speaker recognition, shows that SSC feature coefficients versus the other feature in different SNRs, has the better results. Because of some weakness of LPCC method, the new proposed method, with using SSC feature could cover these weaknesses, therefore combination of SSC and LPCC features will increases the efficiency of the speaker recognition 2.9 percent, and speaker recognition efficiency will be 99.1%.

 

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

Cite: Masood Qarachorloo and Gholamreza Farahani, "New Features to Improve Speaker Recognition Efficiency with Using LPCC and SSC Features," International Journal of Signal Processing Systems, Vol. 4, No. 4, pp. 295-299, August 2016. doi: 10.18178/ijsps.4.4.295-299