GMM Evaluation for Speaker Identification

This work present the results obtained with an automatic speaker identification system, which we have developed and which is based on the GMM speaker modeling method, the identification task is assigned to the GMM-UBM. Several experiments of automatic speaker identification carried out in quiet and noisy environments, on TIMIT database is studied. We experimentally show that increasing the number of adaptation coefficients beyond 10 does not bring a significant improvement of the identification rate in quiet environment. and show degradation on performance when the environment where our identification system is operational becomes noisy.