First International Conference on Spoken Language Processing (ICSLP 90)
In line with the Malaysian Government policy of implementing standard Malay language pronunciation in schools and institutions of higher learning, research on the recognition of spoken Malay language is currently being pursued at the National University of Malaysia.The purpose of this research is to develop a computerised system which can recognize objectively standard Malay language. Uttered Malay words are sampled and broken into phonemes. The voiced and unvoiced phonemes are stored using different parameters. The zero-crossing rate and average magnitude of the signal are used to build the template for the unvoiced phonemes, while the normalized time sampled signal over a selected pitch period and the duration of the phonemes are being used as parameters for the voiced template.In the recognition part, Dynamic Time Warping (DTW) algorithm is used to compare between the voiced template and the input voiced samples. The recognition is done by comparing the sample within the pitch period for the voiced part. However in the case of unvoiced recognition, it is done through the computation of zero-crossing and average magnitude. A database of Malay words is used to compare the recognised words with all the words in the database. A decision as to whether the sound of the uttered word is in accordance with the standard Malay language is achieved through this comparison.Work is actively being undertaken to port the developed system from the IBM PC-AT to the TMS320-based system .
Bibliographic reference. Sharrif, Zainul Abidin Md. / Othman, Masuri / Maiden, Mohammad Ibrahim AKB (1990): "Recognition of standard malaysian language pronunciation", In ICSLP-1990, 277-280.