Cough signal recognition with Gammatone Cepstral Coefficients

Cough Recognition is a valuable classification problem in healthcare. Generally, feature representation contributes a lot to the overall classifying performance. In this paper, a novel feature extraction method, Gammatone Cepstral Coefficients (GTCC), is investigated for cough recognition. The accuracy of GTCC comparing with MFCC is evaluated on a designed cough dataset following a 10 fold cross-validation schemes. Considering the imbalance of that dataset, weighted SVM is applied as the base classifier. The results indicate that GTCC surpass MFCC in modeling cough signals. With combination of GTCC and MFCC, a better performance is achieved. This paper provides a better feature representation prototype in cough recognition.