Can you hear me now? Clinical applications of audio recordings

Audio and speech have several implicit characteristics that have the potential for the identification and quantification of clinical disorders. This PRISMA-guided review is designed to provide an overview of the landscape of automated clinical audio processing to build data-driven predictive models and infer phenotypes of a variety of neuropsychiatric, cardiac, respiratory and other disorders. We detail the important components of this processing workflow, specifically data acquisition and processing, algorithms used and their customization for clinical applications, commonly used tools and software, and benchmarking and evaluation methodologies. Finally, we discuss important open challenges for the field, and potential strategies for addressing them.