New parameters for resolving acoustic confusability between Arabic phonemes in a phonetic HMM recognition system
Iman Abuel Maaly, A R Elobeid, Khandaker Mamun Ahmed · WIT transactions on information and communication technologies · 2002
A Hidden Markov Model (HMM) recognition system is implemented for Arabic Phonemes as units of recognition. An important result of this system is the confision encountered between some phonemes of Arabic (e.g., /h/ (Ha), /?/ (Hamza)), i.e., the recognize could not distinguish between them. New parameters, which are based on a new classification of Arabic phonemes, are added at a higher level of the system for resolving this acoustic confusability and improving the recognition accuracy. These new parameters, which are based on the shape of the tongue and the place of articulation along the vocal tract, are called the Emphatic/nonemphatic, Root and Hamz parameters. The “Emphatic/nonemphatic” parameter gives a performance of 44%, the “Root” parameter gives a performance of 40%, while the “Hamz” parameter could resolve confusibility encountered between the phonemes /h/ and /?/ with a performance of 90%. These new parameters may need accurate estimates of the distances along the vocal tract to improve their performance.