System for characterisation and recognition of Arabic phonemes among Malaysian children using feed-forward neural networks
There has been limited study and research in Arabic phoneme among Malaysians, hence making references to the work and research difficult. Although there have been significant acoustic and phonetic studies on languages such as English, French and Mandarin, to date there are no guidelines or significa...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2012
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/34687/1/NurulAshikinBintiAbdulKadirMFKE2012.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | There has been limited study and research in Arabic phoneme among Malaysians, hence making references to the work and research difficult. Although there have been significant acoustic and phonetic studies on languages such as English, French and Mandarin, to date there are no guidelines or significant findings on Malay language. This study discusses the correct use of Arabic phonemes pronunciation in Malay accent in the simplest form. The International Phonetic Alphabet (IPA) of Arabic chart is used as reference of every recorded speech sample using Malaysian children for sound localisation of every phoneme. Results from Maahad Tahfiz School teachers were used to identify the most suitable samples among the recorded samples. The samples were analysed to determine the origin of each phoneme data by measuring its formant frequencies. Consonants of Standard Arabic (SA) phonemes were studied and an appropriate place of articulation of every phoneme was measured through its formant. Seven out of 25 consonants of SA phonemes of the children’s samples did not give the appropriate formants value were identified. The formant frequency values obtained were used as reference for the database for the proposed recognition system which was developed using Matlab software. All selected samples were randomly divided into 10 disjoint sets of equal size for validation, namely 10-fold cross validation, to estimate the performance of a predictive model. The mean square error (MSE) observed was 0.03, with the speech recognition using a developed neural network (NN) system. The results indicated that the highest training recognition rate obtained for multi-layer and cascade-layer network were 98.8 % and 95.2 % respectively, while the highest testing recognition rate achieved was 92.9 % for both networks and the MSE is 0.04 |
|---|