Acoustic-phonetic approach for automatic evaluation of spoken grammar

Evaluating spoken grammar skills is an important component of evaluating the overall spoken English skills of a candidate. This paper presents a novel approach that incorporates the knowledge of acoustic-phonetics to improve the performance of an existing spoken grammar evaluation technique in a questionanswer paradigm. A novel acoustic parameter, Onset Coherence, to make a one-pass distinction between the onsets of fricatives, stops and vowels is also proposed. The performance of the proposed approach is evaluated against human labeled data obtained from real-life spoken grammar assessment tests. The proposed approach of combining the knowledge from the acoustic-phonetic domain with the information obtained from the automatic speech recognition system leads to an improvement of 8.2% in utterance-level classification.