A New Framework for Automatic Pronunciation Assessment Based on Distinctive Features

This paper presents a novel framework for automatic pronunciation assessment based on distinctive feature analysis. The major idea is to analyze learner’s speech segment to verify whether it conforms to the correct combination of distinctive features. A Distinctive Feature(DF) is a primitive phonetic feature that distinguishes minimal difference of two phones [1]. The overall framework is organized as three layers: DF assessment(DFA) , phone assessment, and continuous speech pronunciation assessment. Various methods can be designed to build DFA modules by extracting suitable acoustic features for each specific DF and classifying the features into score of opposite values. In contrast with conventional method that is based on speech recognition or verification of phonetic units(e.g. phonemes or syllables) , the DFA is language independent and therefore universal. The performance of rudimentary experiments has shown the framework a feasible and compelling new approach.