Reduction and Recognition of Speech Spectra by Computers

Abstract

One of the significant problems of automatic speech recognition has been due to the fact that a speech signal contains redundant information to be processed. In this paper, an algorithm has been developed and employed to reduce the dimensions of quantized patterns by codifying the implicit feature variations. The theory of adaptive pattern recognition has been applied for the recognition of 80-dimensionaI original and 20and 16-dimensional reduced patterns. The results reveal that the indices of correct recognition are unity for all the three sets of patterns, thereby showing that the percentage of correct recognition is one hundred percent not only for the original 80-dimensionaI patterns but also for the 20and 16-dimensional reduced patterns.

Indexing Terms:

• Automatic Speech Recognition
• Adaptive Pattern Recognition