Self-organization map of spiking neurons evaluation in phoneme classification

The Self-organizing Maps, proposed by Kohonen, have been used with a great deal of success in many applications. However, the basic SOM is designed to map patterns, or feature vectors, from an input space into an output space and does not take time into account. Speech recognition is a sequence of acoustic information and can´t be considered as a static vector of information. In order to classify temporal sequences we present in this paper how to use spiking neurons for unsupervised competitive learning, preserving self-organizing maps. In fact we present two variants of SOM, the Leaky Integrators Neurons and the Spiking_SOM models. The Leaky Integrators Neurons model is based on the conservation of information, which makes it possible to consider the temporal order between the successive samples by using a mechanism called Leaky Integrators and the Spiking_SOM model preserves traditional model SOM, however it represents the characteristic to modify the learning function. The case study of the proposed models is phoneme classification in continuous speech.