Improved Self-Organizing Maps and Speech Compression

Recent developments of Self-Organizing Maps or Kohonen networks become more and more interesting in many fields such as: pattern recognition, clustering, speech recognition, data compression, medical diagnosis... Kohonen networks is unsupervised learning models. The results obtained by the Kohonen networks are dependent on their parameters such as the architecture of the Kohonen map, the later has a great impact on the convergence of learning methods. The selection of the architecture of Kohonen networks, associated with a given problem, is one of the most important research problems in the neural network research. In this paper, we model this problem of neural architecture in terms of a mixed-integer non linear problem with linear constraints. To solve this model of optimization for the network architectures, we propose the genetic algorithm. Also, we implemented and evaluated the proposed method and speech compression algorithms. Speech compression is the technology of converting human speech into an efficiently encoded representation that can later be decoded to produce a close approximation of the original signal. The numerical results demonstrated the effectiveness of the new model.