423. Detecting Pertussis Using Voice Recognition Technology
Session: Poster Abstract Session: Novel Devices and Technologies
Thursday, October 18, 2012
Room: SDCC Poster Hall F-H
Background: Pertussis is highly contagious, thus, prompt identification of cases is essential to control outbreaks. Early stages of the disease resemble other respiratory infections. In many cases, the cough gradually becomes more severe, and in children a paraoxysmal phase begins. This stage is characterized by bursts of rapid coughs followed by gasps and the characteristic whoop sound. Clinicians experienced with the disease can easily identify these sounds. However, many clinicians have never seen a case, and thus may miss initial cases during an outbreak. The purpose of this project was to use voice-recognition software to distinguish pertussis coughs from croup and other coughs.

Methods: We collected a series of sound files representing pertussis cases available on the Internet on YouTube and public health websites. We also collected multiple sound files available on YouTube representing croup and miscellaneous coughing by children who did not have pertussis. We manually categorized coughs as either pertussis or non-pertussis, and extracted features for each category. The backgrounds for coughs are different (speech, noise, cry, clean). We used Mel-frequency cepstral coefficients (MFCC), a sampling rate of 16 KHz, a frame Duration of 25 msec, and a frame rate of 10 msec. The coughs were passed through a filter set to pass frequencies between 5 Hz and 6 KHz.  Each cough is divided into 3 sections of proportion 3-4-3. The average of the 13 MFCCs for each section is computed and made into a 39-element feature vector used for the classification. The following machine learning algorithms were implemented: Neural Networks, K-Nearest Neighbor (KNN), and a 200 tree Random Forest (RF). Data was reserved for cross-validation of the KNN and RF and the Neural Network was trained 100 times and the averaged results are presented.

Results: After categorization, we have 16 examples of non-pertussis coughs and 31 examples of pertussis coughs. Over 90% of all pertussis coughs were properly classified as pertussis.  The error rates were: Type I error 3.7%, 4.3%, and 8.5%; Type II error 4.3%, 0.0 %, 0.0% using the Neural Network, Random Forest, and KNN, respectively.

Conclusion: Our results suggest that we can build a robust classifier to assist clinicians and the public to help identify pertussis cases in children presenting with typical paroxysmal symptoms.

Danny Parker, PhD1, Joseph Picone, PhD2, Amir Harati, MS2, Shuang Lu2, Marion H. Jenkyns3 and Philip M. Polgreen, MD4, (1)GTD Unlimted, Oxford, MS, (2)Temple University, Philadelphia, PA, (3)Oxford High School, Oxford, United Kingdom, (4)University of Iowa, Iowa City, IA


D. Parker, None

J. Picone, None

A. Harati, None

S. Lu, None

M. H. Jenkyns, None

P. M. Polgreen, None

Previous Abstract | Next Abstract >>

Findings in the abstracts are embargoed until 12:01 a.m. PST, Oct. 17th with the exception of research findings presented at the IDWeek press conferences.