Pharyngeal and emphatic sounds in arabic speech recognition
Yousef Ajami Alotaibi, Mahyar Shahsavari · 1997
The Arabic language is a Semitic language which has many differences when compared to European languages such as English. One of these differences is the existence of pharyngeal and emphatic sounds that represent a substantial part of Arabic phonemes. This study concentrated on finding the unique characteristics of these sounds, comparing these sounds to English sounds, and investigating the effect of these sounds on an Arabic speech recognition system. It was found that some pharyngeal and emphatic sounds are unique and it is easy for the recognizer to identify them correctly. On the other hand, there are other sounds that are very similar and they degraded the Arabic speech recognizer. This research also showed that qaaf sound, /q/, created many problems in the Arabic speech recognition process because it has the same features as many other Arabic sounds. This research studied some of the speech recognition algorithms with the concentration on phanageal and emphatic sound sets. Performance of several distance measures were compared and evaluated. It was concluded that the likelihood ratio distance measure achieved the best performance; on the other hand, the Euclidean distance measure has the worst performance among the five distance measures studied. The performance of two training methods was investigated and compared with the presence of Arabic pharyngeal and emphatic sounds. An Arabic speech recognition system based on the pattern recognition model was developed. Pharyngeal and emphatic sounds were extracted from their carrier words and they were investigated solely without the effect of their neighboring sounds in the same word.