Voice Quality Assessment by Simulating GRBAS Scoring

Abstract

This paper is about the assessment of voice quality as required routinely in hospital voice clinics. It describes a computer application capable of analysing recordings of a patient's voice and producing quantitative assessments of its quality, simulating those traditionally made by trained speech and language therapists (SLTs). Adopting a machine learning approach based on a database of recordings and assessments by a team of SLTs required measurements of consistency to be taken into account. The means of doing this, details of the machine learning approaches and the performance of the resulting algorithms are presented.