Fluidity: Real-time feedback on acoustic measures of second language speech fluency

Fluency development is a primary goal of most second language learners. This paper describes Fluidity, a Java application that is designed to help second language learners develop their speech fluency through various practice functions and with unique, avatar-based feedback given in real-time. Learners may take advantage of several practice options including scripted and free speech practice. While users speak, the application records their speech, analyzes it in real-time with respect to utterance fluency features, and provides feedback via facial expressions of the on-screen avatar. Learners may also review their practice afterward with several visualizations of their fluency. This paper describes proof-of-concept testing of the fluency feature detection mechanism using existing recordings of spontaneous speech. Results show that the application measures various fluency features (e.g., syllable count, silent pause count) at an accuracy comparable to or better than a commonly used offline method. It also has a high correlation with manual corpus measurements on several fluency measures.