Age and gender recognition issue provides important capability in various processing areas dealing with telephone speech systems to consider the identity of an individual using the recorded voice content. In this paper, a new age/gender recognition system is proposed based on the learned models using the sparse representation of wavelet packet coefficients in different decomposition levels. The proposed classification approach includes a learning step to provide related atoms to each signal class and test step to evaluate the performance of the classification scheme. The dictionary atoms are trained over male and female speaker data using sparsity and coherence constraints. The experimental results show that the presented algorithm obtains better results than the earlier methods in this context and also in the presence of background white noise.
