Formant based linear prediction coefficients for speaker identification

Here Formant Based Linear Prediction Coefficient (FBLPC) features are proposed for speaker identification for all environments. Gaussian Mixture Models (GMMs) are used for classification of speakers. The identification performance of Linear Prediction Coefficient (LPC) features is computed and compared with the identification performance of FBLPC features. The performance of FBLPC features is found superior in clean and noisy environments. A database of fifty speakers for ten Hindi digits is used for simulation of results.