New variant of the Self Organizing Map in Pulsed Neural Networks to Improve Phoneme Recognition in Continuous Speech

Speech recognition has gradually improved over the years, phoneme recognition in particular. Phoneme recognition plays very important role in speech processing. Phoneme strings are basic representation for automatic language recognition and it is proved that language recognition results are highly correlated with phoneme recognition results. Nowadays, many recognizers are based on Artificial neural networks have been applied successfully in speech recognition applications including multi-layer perceptrons, time delay neural network, recurrent neural network and self-organizing maps (SOM), but present some weaknesses if patterns involve a temporal component. Let's note for example in speech recognition or contextual information, where different of the time interval, is crucial for comprehension. In this paper, we propose a new variant SOM made of spiking neurons, with a view to emphasising the temporal aspect of the data which might serve as an input, in order to improve phoneme classification accuracy. The proposed variant, the Leaky Integrators Neurons, is like the basic SOM, however it represents the characteristic to modify the learning function and the choice of the best matching unit (BMU). The proposed SOM variant, show good robustness and high phoneme classification rates.