Arabic/English automatic spoken language identification

Arabic/English language identification system for short duration, telephone quality speech utterances based on speaker identification, vowel spotting, and continuous hidden Markov models is presented. The effect of vowel spotting on both speaker identification and language identification accuracy is also studied. Three different designs and implementations of this system are introduced in this study, the first one is based on speaker identification, the second is based on speaker identification using a language dependent vowel spotter while the third one is based on speaker identification using a language independent vowel spotter. Tests on the three systems are conducted and their performance are analyzed and a comparison between them is achieved. Methods to improve the performances are also suggested.