In the context of content-based multimedia indexing gender identification based on speech signal is an important task. In this paper a set of acoustic and pitch features along with different classifiers are compared for the problem of gender identification. We show that the fusion of features and classifiers performs better than any individual classifier. Based on such conclusions we built a system for gender identification in multimedia applications. The system uses a set of Neural Networks with acoustic and Pitch related features. 90% of classification accuracy is obtained for 1 second segments and with independence to the language and the channel of the speech. Practical considerations, such as the continuity of speech and the use of mixture of experts instead of one single expert are shown to improve the classification accuracy to 93%. When used on a subset of the Switchboard database, the classification accuracy attains 98.5% for 5 seconds segments.