Genetic algorithms and fuzzy approach to Gaussian mixture model for speaker recognition

Gaussian mixture models (GMMs) is currently the most popular approach to speaker recognition. In speaker recognition, the major problem is how to generate a set of the GMM for identification purposes based upon the training data. Due to the hill-climbing characteristic of the expectation maximization (EM) method, it is sensitive to the initial model parameters and easy to lead to a sub-optimal model in practice. To resolve this problem, this paper proposes a new hybrid training method based on the global searching capability of genetic algorithms (GA) and the effectiveness of fuzzy approach to obtain GMMs with optimized model parameters. Experimental results based on PKU-SRSC database showed that this method could obtain more optimized GMMs and better results than the hybrid GA based on the EM re-estimation and the traditional EM method.