The problem of pronunciation evaluation of sentences is defined as the combination of word based subjective pronunciation scores. The mean subjective word score criterion is proposed and modeled with the combination of word-based objective assessment. The word objective metric requires no a priori studies of common mistakes, and it makes use of class based language models to incorporate wrong and correct pronunciations. Wrong pronunciations are automatically generated by employing competitive lexicon, and students' native language phonetic rules. Subjective-objective sentence score correlations greater than 0.5 can be achieved when the proposed sentence based pronunciation criterion is approximated with the combinations of word-based scores. Finally, the subjective-objective sentence score correlations reported here are very comparable with those published elsewhere with methods that require a priori studies of pronunciation errors.

Index Terms: Computer-aided pronunciation training, subjective criterion, second language learning, ASR