Communication-Oriented Automatic Assessment System for Accented Spoken Chinese in Read-Aloud Tasks

The development of speech signal processing and deep learning has brought in many intelligent language learning tools. However, non-native Chinese learners (second-language or L2 learners) are often discouraged by language assessment applications on the market because of their accent. By contrast to artificial models, human experts usually give a higher score based on the performance of L2 learners in read-aloud tasks. In order to precisely assess the spoken Chinese of L2 learners in communication-oriented environments, we design a new assessment system, AsAsC (short for Assessment System for Accented Spoken Chinese) featuring comprehensive indexes and feasible quantitative schemes. Experiments on the real-world dataset show that AsAsC achieves a more human-like assessment capability in comparison to three typical enterprises’ assessment systems.