Speaker-independent speech recognition system using Kohonen self-organizing feature map

Date of Publication

1999

Document Type

Bachelor's Thesis

Degree Name

Bachelor of Science in Electronics and Communications Engineering

Subject Categories

Engineering | Systems and Communications

College

Gokongwei College of Engineering

Department/Unit

Electronics and Communications Engineering

Abstract/Summary

In the past few years, there has been much noteworthy advancement in artificial neural networks. One such classification of a neural network model was presented by Teuvo Kohonen, which produces what he calls self-organizing feature maps (SOFM) similar to how the brain works. The goal of the SOFM algorithm is to transform an incoming signal pattern of arbitrary dimensions into a discrete map, and to perform this transformatoin adaptively in a topologically ordered fashion.

This pattern classification ability of SOFM is explored for a practical speech recognition problem in this project. This thesis aims to develop a system, using Kohonen's SOFM algorithm, to recognize single word utterances independent of the speaker. With the proper algorithm and training, the SOFM forms a clustering of the inputs to perform word recognition.

The speaker-independent speech recognition system accepts as input isolated words stored as digital speech files. The speech files are preprocessed in order to extract the LPC coefficients of each file, which will serve as the input to the neural network. The SOFM is used to create a topological map of the commands in an unsupervised fashion. Once a topological map is generated, fine-turning is done using Optimum Learning Vector Quantization 1 (OLVQ1) algorithm. An architectural structure of the final map is designed using VHDL software. The design implements the Manhattan Distance computation using the IEEE format on real numbers.

The system achieved a recognition rate of 97.5%.

Abstract Format

html

Language

English

Format

Print

Accession Number

TU09303

Shelf Location

Archives, The Learning Commons, 12F, Henry Sy Sr. Hall

Physical Description

1 vol. (various pagings) ; Computer print-out.

Keywords

Automatic speech recognition; Pattern recognition systems; Speech processing systems.

This document is currently not available here.

Share

COinS