8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


An Acquisition Model of Speech Perception with Considerations of Temporal Information

Ching-Pong Au

City University of Hong Kong, China

Speech Perception of humans begins to develop as young as 6- month-old or even earlier. The development of perception was suggested to be a self-organizing process driven by the linguistic environment to the infants [1]. Self-organizing maps have been widely used for modeling the perception development of infants [2]. However, in these models, temporal information within speech is ignored. Only single vowels or phones have little variations along time can be represented in this kind of models. In the present model, temporal information of speech can be captured by the self-feeding input preprocessors so that the sequence of speech components can be learnt by the self-organizing map. The acquisition of both the single vowels and diphthongs will be demonstrated in this paper.

Full Paper

Bibliographic reference.  Au, Ching-Pong (2003): "An acquisition model of speech perception with considerations of temporal information", In EUROSPEECH-2003, 2133-2136.