A tablet application for English pronunciation assessment

Automatic pronunciation scoring is becoming an interesting field of study as on-line tools and mobile applications are becoming more popular. In this paper, we will investigate research and development of a pronunciation scoring tool for American English designed for tablets. We have investigated mainly three dimensions in pronunciation namely phonetic articulation, intonation and timing. We worked with real data recorded from mobile devices in real environments. We analyzed 527 recordings from 100 different speakers with varying degrees of knowledge of American English and with different first language backgrounds. We developed a scoring algorithm for the pronunciations at the sentence and word levels and then compared our results with human ratings. We have found a correlation value of 0.73 for agreement between our proposed method and human ratings while the inter-labeler agreement was 0.83. We also asked the opinion of 10 subjects regarding the user interface and ease of use for the feedback given in three dimensions. Average user satisfaction for the application was 4.2/5.