InterSpeech 2021

The Four-way Classification of Stops with Voicing and Aspiration for Non-native Speech Evaluation
(3 minutes introduction)

Titas Chakraborty (IIT Bombay, India), Vaishali Patil (IIIT Pune, India), Preeti Rao (IIT Bombay, India)
The four-way distinction of plosives in terms of voicing and aspiration is rare in the world’s languages, but is an important characteristic of the Indo-Aryan language family. Both perception and production pose challenges to the language learner whose native tongue does not afford the specific distinctions. A study of the acoustic-phonetics of the sounds and their possible dependence on speaker characteristics, such as gender or native tongue, can inform methods for accurate feedback on the quality of the phones produced by a non-native learner. We present a system for the four-way classification of stops building on features previously proposed for aspiration detection in unvoiced and voiced plosives. Trained on an available dataset of Hindi speech by native speakers, the system works reliably on production data comprising Bangla words uttered by native Bangla and non-native (American English L1) speakers. The latter display a variety of articulation patterns for the given target contrasts, providing useful insights related to L1 influence on the voicing-aspiration production in word-initial CV contexts.