ISCA International Workshop on Speech and Language Technology in Education (SLaTE 2009)
Wroxall Abbey Estate, Warwickshire, England
Recently, in the language testing field, automatic speech recognition (ASR) technology has been used to automatically score speaking tests. This paper investigates the impact of audio quality on ASR-based automatic speaking assessment. Using the read speech data in the International English Speaking Test (IEST) practice test, we annotated audio quality and compared scores rated by humans, speech recognition accuracy, and the quality of features used for the automatic assessment under high and low audio quality conditions. Our investigation suggests that human raters can cope with low-quality audio files well, but speech recognition and the features extracted for the automatic assessment perform worse on the low audio quality condition.
Bibliographic reference. Chen, Lei (2009): "Audio quality issue for automatic speech assessment", In SLaTE-2009, 97-100.