Adding quiet and music detection capabilities to FDA-approved cochlear implant research platform

摘要

This paper presents the addition of quiet and music detection capabilities to our previously developed environment-adaptive cochlear implant speech processing pipeline. A modified Voice Activity Detector is proposed which provides quiet frame detection in addition to voice and no-voice activity detection in a computationally efficient manner. Music detection is achieved via a two-class Gaussian mixture model classifier requiring no extra computation for feature extraction. The results obtained indicate this added capability generates high detection rates for both quiet and music conditions.