Exploratory data analysis on nuclei in cantonese dysarthric speech

Abstract

Phone substitutions, distortions, deletions, and insertions are some of the main problems in dysarthric speech. This work aims to explore the articulatory error patterns in dysarthric speech, which provides insights for the improvement of automatic dysarthric speech analysis technologies. A set of dysarthric speech is collected and phonetically transcribed manually by different transcribers. Transcriptions are mapped into distinctive feature values. Error rates for each value are extracted. Substitutions and distortions are found to be the major errors in dysarthric speech. Their error patterns are analyzed in this paper. The analysis may provide guidance for labelling dysarthric speech errors, which will be useful to future development of technologies to achieve automated analysis of dysarthric speech.