Evaluating Vowel Pronunciation in Computer Assisted Pronunciation Training.

Abstract

Computer Assisted Pronunciation Training (CAPT) applications are tools that can be used when learning a second language. By evaluating the speech of a student, the CAPT system is able to give automatic feedback on his or her pronunciation performance.Two important properties in Norwegian pronunciation is the quantity and quality of the vowels. It is therefore important that students get to practice this.Feedback was produced by an ASR based CAPT system, where a speech recognizer evaluated the pronunciation produced by different speakers. However, sind ASR is prone to errors, verification was later performed to test the correctness of the recognizers results.The recognizer had an error-rate of 7.5 % when evaluating vowel quantity, and an error rate of 42.1 % when evaluating vowel quality. After verification, the first error rate was reduced to 1.35 % by rejecting 7.2 % of the results. The second error rate was reduced to 27,7 % by rejecting 23.5 of the results.The use of such a system could therefore be justified for evaluating the vowel quantity in the pronunciation, but not vowel quality.