Neural Network Model of Speech Dynamics



The paper addressed the problem of modeling the dynamics of speech from the perspective of self-organizing neural networks. In this context, the paper started from the premise that the dynamic behavior of the phonetic constituents can be identified in natural speech at the neural level and investigated the possibility of extracting the features of phones/phonemes using temporal self-organizing maps (SOMs). A variant of temporal SOM was proposed that proved able to encompass the phones/phonemes dynamic behavior and also model the feature trajectory traces as they appear in the feature space. The results of simulations revealed the potential offered by this kind of temporal SOMs to extract the dynamic features and model the trajectory for any individual phone/phoneme. The model appeared consistent with recent evidences in the field of neurodynamics, and can be extended as well for modeling brain dynamics in language processing.


Neural networks, Self-Organizing maps, Semantic modeling, Time series modeling, Speech processing.


Full Text:



  • There are currently no refbacks.