الفهرس | Only 14 pages are availabe for public view |
Abstract Syllables are the fundamental units of Arabic language. The proposed 2Neural Network based Arabic Speech Segmentation System (NNASS)3 is an adaptive Arabic speech syllable boundaries identifier that mainly serves as an automatic segmentation tool for speaker independent 2Arabic speech verification (ASV)3 and speech corpus/database construction systems. Cpestral peaks extracted from recorded speech signal within a certain validation thresholds assignment are considered probable boundaries. These probable boundaries are applied to NNASS to classify them into valid or invalid ones. An algorithm using neural networks is developed to train the features of valid boundaries/ cores. A program is developed to precisely identify the boundaries/cores from the test utterance, where the segmentation is done at their locations. The accuracy of NNASS was 87 % and 92.2 % identification rates with a semi-automatic labeling of the test dataset for verification within 10 and 20 milliseconds using two sample sizes. It will be shown that the system can be expanded to include more trained utterances for more than application |