Detecting symptoms of diseases in poultry through audio signal processing

Abstract

We developed an audio signal processing algorithm that detects rales (gurgling noises that are a distinct symptom of common respiratory diseases in poultry). We derived features from the audio by calculating mel frequency cepstral coefficients (MFCCs), clustering the MFCC vectors, and examining the distribution of cluster indices over a window of time. The features are classified with a C4.5 decision tree. Our training data consisted of eight minutes of manually labeled audio selected from 25 days of continuous recording from a controlled study. The experiment group was challenged with the infectious bronchitis virus and became sick, while the control group remained healthy. We tested the algorithm on the entire dataset and obtained results that match the course of the disease. Algorithms such as this could be used to continuously monitor chickens in commercial poultry farms, providing an early warning system that could significantly reduce the costs incurred from disease.