Assessing automatic speech recognition in measuring speech intelligibility: A study of malay speakers with speech impairments

This paper investigates the potential of Automatic Speech Recognition application in measuring the speech intelligibility of speech impaired speakers. ASR has been suggested to be a useful way of intelligibility measures in the literature. In this research, we tested ASR's performance in measuring speech intelligibility in the context of speakers with speech impairments. A speaker independent ASR system was developed to produce the intelligibility scores automatically. A panel of Speech Language Pathologists involved in assessing the speech intelligibility subjectively to provide the benchmark scores. Results show that the correlation between ASR with SLPs subjective scores is significant at p<;0.0f, r(0.57) = 0.1. This indicates a positive correlation of ASR automatic measurement with human subjective rating, which the WRA increase with the increment of intelligibility scores.