I-Vectors is a feature extraction technique which is predominantly used for speaker identification and recognition. We decided to use this technique to help us detect Dysphonic patients from Normal patients with higher levels of accuracy. We used Gaussian-Mixture Models for helping us extract I-Vectors while the final classification was done using Support Vector Machines (SVM). Using I-Vectors we achieved accuracy of 98%, significantly higher than without using I-Vectors.
This paper is a comparative study of how Support Vector Machines (SVM), Naive Bayes and K-Nearest Neighbors (KNN) can be used for classifying three voice disorders - Dysphonia, Vocal Fold Paralysis and Laryngitis along with normal speech samples. For our study, we made use of the proposed I-Vector Feature extraction technique.