Preliminary Identification of Speech Hoarseness from Cancer and Normal Speech

The problem of speech hoarseness already arise for long especially in any kind of customer service interaction, in education or audio medical studies. But this issue is not widely in cover or explore especially in terms tendency checking for variety of video, in gender, age, and between normal or hoarseness speech model. The current studies conducted initial experiments with audio speech signal contains of normal and hoarseness data from `Saarbrücken Voice Database'. This study is prevalence for investigating the difference between normal and hoarseness voice speech by implementing classification algorithm of K-Nearest Neighbor(KNN) and Gaussian Mixture Model(GMM) in order to find the best algorithm in identifying the different datasets of audio and the performance for different types of algorithm. For KNN, the classification accuracies achieved 70% for both combination of datasets while with GMM achieved higher accuracies for 90% from the overall combination. Through this preliminary studies can embarked more focus in identifying on hoarseness cases for education fields and not only to medical sector.