Multimodal biometrics using multiple feature representations to speaker identification system

Multimodal biometrics combines information coming from multiple biometrics with a key objective to reduce the limitations associated with any single biometric method such as low accuracy, limited security, noisy measurements, etc. In this study, different multimodal speaker identification approaches are investigated. Linear predictive coding features, Mel-frequency cepstral coefficients features, discrete wavelet based linear predictive coding features are examined with the use of different combinations of features applied to the identification system. In building the multimodal system, fusion is realized at the score level using Gaussian mixture model. The system is tested on publicly available data set and shows improvement in the classification rate for all feature extraction methods.