Developing and Testing Audio Data Processing Modules in Python to Connect to and Data Be Scored by ASS Cloud Server

Abstract

Automatic Speech Scoring (ASS) system developed on a basis of automatic speech recognition (ASR) technology is a powerful computer-assistant tool for oral test scoring. However, due to the limits of high equipment costs and high-tech operating costs of a local ASS, ASS cloud services have become the first choice of most oral English teachers and learners. The purpose of this paper is to develop and test modules in Python to preprocess the audio data, connect to the cloud server, and convert JSON data format into common Excel form. 1056 pieces of audio data were collected from test-takers’ read-aloud task of CEST-4 (College English Speaking Test band 4)) and six variables (i.e., “pronunciation”, “fluency”, “integrity”, “speed”, “duration”, and “overall”) were defined. After analyzing the data of the test results, it is found that the oral test score is mostly affected by the “pronunciation” and “integrity”, and the accuracy of pronunciation is the strongest predictor of oral performance. The modules and functions are helpful for teachers and students to use in daily oral test/practice, and these modules can also be employed in other second language oral test scored by ASS cloud sever, like oral Chinese test. Our results can provide reference and guidance for future oral research and teaching.