An Implementation of Speech Recognition System based on Developmental Approach

Automatic Speech Recognition (ASR) is an important but difficult problem. Although there have been developed several different approaches in the past decades, the recognition accuracy is still not high enough. Recently, an autonomous development approach has been proposed and drawn more and more attention in the field of robotics and artificial intelligence. It is interesting and worth investigating that how to utilize this idea in the solution for ASR. In this paper, we give a detail discussion on an implementation of speech recognition system based on the developmental approach. Most of our idea is derived from the excellent work by Zhang and Weng. We first give a brief overview of ASR and autonomous development approach, and then several complementarities and detail algorithms are discussed such as using Self Organizing Map (SOM) in the Sensory Mapping module and Simple Recurrent Network in the Cognitive Mapping module. Last, we partially repeat the experiments in Zhang’s work on our own platforms, and give comparison results by using SOM, IPCA, and manually selected features.