Development and Evaluation of a Speech-to-Noise Ratio Feedback System

Abstract

People with Parkinson's disease (PD) often present reduced vocal loudness, which may impact the intelligibility of their speech especially in noisy environments. To address this issue, we developed an assistive speech-to-noise ratio feedback (SNF) system that estimates the user's speech signal-to-noise ratio (SNR) in real-time and activates an audible alarm if the SNR falls below a predefined threshold. The proposed SNF system is comprised of a pair of over-the-ear binaural microphones for audio data acquisition, and a mobile application (app) that implements the key algorithms for coherence-based own voice detection (OVD), speech SNR estimation, and alarm triggering. The proposed SNF system performance was evaluated through electroacoustic and subjective tests under a variety of environ-mental conditions. Our results indicated that the lightweight OVD algorithm effectively differentiated the user's own voice from other audio signals when appropriate thresholds were set. The subjective-testing results also demonstrated that, during use, the SNF system effectively increased users' speech intensity.