Feature enhancement and selection methods for isolated Malay speech recognition

Automatic speech recognition (ASR) is a technique to translate automatically incoming speech signal into their contextual information. In the pass few decade, various acoustic feature extraction and classification algorithms have been developed for native English speech recognition and different lan...

Full description

Saved in:

Bibliographic Details
Format:	Thesis
Language:	English
Subjects:	Automatic speech recognition Speech perception Malay language
Online Access:	http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/1/Page%201-24.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/2/Full%20text.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/4/Chong%20Yen%20Fook.pdf
Tags:	Add Tag No Tags, Be the first to tag this record!

id	my-unimap-77999
record_format	uketd_dc
spelling	my-unimap-779992023-03-06T04:27:59Z Feature enhancement and selection methods for isolated Malay speech recognition Vikneswaran, Vijean, Dr. Automatic speech recognition (ASR) is a technique to translate automatically incoming speech signal into their contextual information. In the pass few decade, various acoustic feature extraction and classification algorithms have been developed for native English speech recognition and different languages spoken around the world using acoustic signals. Research in Automatic Speech Recognition (ASR) by machines had been done for more than five decades. Various research findings have been reported in recent years in speech recognition for many different languages. However, every languages having their own unique words structure. As examples, English words are formed due to the changes of phoneme in the based word itself according to its group of words and Malay words allow addition of affixes to the base word to form new words. In this research, signal processing techniques are applied to the acoustic signals in an effort to recognize the Malay speech. To reduce the misclassification, the recorded speech signals were segmented to remove the unvoiced speech (noise). In this research works, parametric Linear Prediction Coefficients (LPC), Linear Prediction Cepstral Coefficient (LPCC), Weighted Linear Prediction Coefficients (WLPCC), Mel-Frequency Cepstral Coefficients (MFCC) and non-parametric Wavelet Packet Transform based Energy and Entropy (WPT-EE) representations of features were extracted. The features extracted were enhanced to increase the discriminant ability using artificial bee colony based clustering. Then, the enhanced features set were dimensionally reduced by using two feature selection techniques. They are binary particle swarm optimization (BPSO) and discrete artificial bee colony (DABC) feature selection technique. Last, two classifiers as the probabilistic neural network (PNN) and extreme learning machine (ELM) were used to evaluate the performance of extracted and enhanced features from recorded Malay speech signal. The proposed artificial bee colony based feature enhancement (ABC-FE) features show promising average results of 99.61% (Speaker Dependent) and 96.21% (Speaker Independent). Experimental results showed that the average accuracy obtained by using hybrid features of LPC, LPCC, WLPCC, MFCC and WPT-EE for Speaker Dependent and Speaker Independent with ELM classifier were 97.89% (PSO)- 98 features and 99.33% (ABC)-67 features for Malay speech recognition. Universiti Malaysia Perlis (UniMAP) Thesis en http://dspace.unimap.edu.my:80/xmlui/handle/123456789/77999 http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/3/license.txt 8a4605be74aa9ea9d79846c1fba20a33 http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/1/Page%201-24.pdf fbcdebec6d2d2fcda16c9e688ffd43cb http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/2/Full%20text.pdf 00c585d3bc6a49407828c891f5bc9ce7 http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/4/Chong%20Yen%20Fook.pdf 40adb4bab582b0c1dfadb2c5d77aa4e0 Universiti Malaysia Perlis (UniMAP) Automatic speech recognition Speech perception Malay language School of Mechatronic Engineering
institution	Universiti Malaysia Perlis
collection	UniMAP Institutional Repository
language	English
advisor	Vikneswaran, Vijean, Dr.
topic	Automatic speech recognition Speech perception Malay language
spellingShingle	Automatic speech recognition Speech perception Malay language Feature enhancement and selection methods for isolated Malay speech recognition
description	Automatic speech recognition (ASR) is a technique to translate automatically incoming speech signal into their contextual information. In the pass few decade, various acoustic feature extraction and classification algorithms have been developed for native English speech recognition and different languages spoken around the world using acoustic signals. Research in Automatic Speech Recognition (ASR) by machines had been done for more than five decades. Various research findings have been reported in recent years in speech recognition for many different languages. However, every languages having their own unique words structure. As examples, English words are formed due to the changes of phoneme in the based word itself according to its group of words and Malay words allow addition of affixes to the base word to form new words. In this research, signal processing techniques are applied to the acoustic signals in an effort to recognize the Malay speech. To reduce the misclassification, the recorded speech signals were segmented to remove the unvoiced speech (noise). In this research works, parametric Linear Prediction Coefficients (LPC), Linear Prediction Cepstral Coefficient (LPCC), Weighted Linear Prediction Coefficients (WLPCC), Mel-Frequency Cepstral Coefficients (MFCC) and non-parametric Wavelet Packet Transform based Energy and Entropy (WPT-EE) representations of features were extracted. The features extracted were enhanced to increase the discriminant ability using artificial bee colony based clustering. Then, the enhanced features set were dimensionally reduced by using two feature selection techniques. They are binary particle swarm optimization (BPSO) and discrete artificial bee colony (DABC) feature selection technique. Last, two classifiers as the probabilistic neural network (PNN) and extreme learning machine (ELM) were used to evaluate the performance of extracted and enhanced features from recorded Malay speech signal. The proposed artificial bee colony based feature enhancement (ABC-FE) features show promising average results of 99.61% (Speaker Dependent) and 96.21% (Speaker Independent). Experimental results showed that the average accuracy obtained by using hybrid features of LPC, LPCC, WLPCC, MFCC and WPT-EE for Speaker Dependent and Speaker Independent with ELM classifier were 97.89% (PSO)- 98 features and 99.33% (ABC)-67 features for Malay speech recognition.
format	Thesis
title	Feature enhancement and selection methods for isolated Malay speech recognition
title_short	Feature enhancement and selection methods for isolated Malay speech recognition
title_full	Feature enhancement and selection methods for isolated Malay speech recognition
title_fullStr	Feature enhancement and selection methods for isolated Malay speech recognition
title_full_unstemmed	Feature enhancement and selection methods for isolated Malay speech recognition
title_sort	feature enhancement and selection methods for isolated malay speech recognition
granting_institution	Universiti Malaysia Perlis (UniMAP)
granting_department	School of Mechatronic Engineering
url	http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/1/Page%201-24.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/2/Full%20text.pdf http://dspace.unimap.edu.my:80/xmlui/bitstream/123456789/77999/4/Chong%20Yen%20Fook.pdf
_version_	1776104259320807424

Feature enhancement and selection methods for isolated Malay speech recognition

Similar Items