Probabilistic cooperation of connectionist expect modules: validation on a speaker identification task

The author presents and evaluates a modular connectionist system for speaker identification. Modularity has emerged as a powerful technique for reducing the complexity of connectionist systems and allowing prior knowledge to be incorporated into their design. Thus, for systems where the amount of training data is limited, modular systems incorporating prior knowledge are likely to generalize significantly better than a monolithic connectionist system. An architecture is developed which achieves speaker identification based on the cooperation of several connectionist expert modules. When tested on a population of 102 speakers extracted from the DARPA-TIMIT database, perfect identification was observed. In a specific comparison with a system based on multivariate autoregressive models, the modular connectionist approach was found to be significantly better in terms of both identification accuracy and speed.<>