Analyzing Speech and Pronunciation Performance Virtually
Abstract: For the past several years we have been creating a system using a motion-sensing camera. The IPA [Interview and Presentation Assistant] combined facial, speech, and gestural recognition and helped students get personalized feedback based on parameters set by human input. However, while we were doing this, voice recognition systems such as Google, Alexa and Siri started to get really, really good. Because of this, we changed the direction. We are currently developing a system to evaluate speech performance of EFL students in Japan which focuses on the pronunciation of individual sounds, word/sentence stress, and rhythm/intonation. Teachers input a score from an audio clip recorded by the student. This scores are then fed into the system to create an algorithm which in turn can reproduce the teacher’s score to a certain degree without the teacher being present. This research is in its initial stage, but the results show the possibility to create a system which will help students improve their English speaking ability. This will help provide the important one-to-one personalized feedback they need to do so.
Presider: Douglass Scott, Waseda University