Monitoring voice condition using smartphones

Abstract

Smartphone mediated voice monitoring has the potential to support voice care by facilitating data collection, analysis and biofeedback. To field-test this approach we have developed a smartphone app that allows recording of voice samples alongside voice self-report data. Our longterm aim is convenient and accessible voice monitoring to prevent voice problems and disorders. Our current study focussed on the automatic detection of voice changes in healthy voices that result from common transient illnesses like colds. We have recorded a database of approximately 700 voice samples from 62 speakers and selected a subset of 225 voice samples from 8 speakers who had submitted at least 10 recordings and reported at least one instance of a moderate cold. We extracted 12 acoustic parameters and applied multivariate statistical process control procedures (Hotelling's T2) to detect whether instances of cold caused violations of distributional control limits. Results showed significant association between control limit violations and reporting of a cold. While there is scope for further improvement of sensitivity and specificity of the procedure, it could already support early detection of voice problems, especially if mediated by voice experts.