MPEG-7 audio features based voice pathology detection

Voice pathology detection based on MPEG-7 low-level audio feature is proposed in this paper. MPEG-7 features contain both audio and video descriptors, and originally were introduced for multimedia indexing. Indexing is related to event detection, and because the pathological voice is a separate event than the normal voice, we use MPEG-7 audio descriptors as a tool to detect voice pathology. In the proposed voice pathology detection, MPEG-7 low level audio descriptors are extracted from input voice, and support vector machine (SVM) is used as classifier. Fisher discrimination ratio (FDR) is applied on the extracted descriptors to identify the most significant features for the detection. The experimental results on the MEEI database show that the proposed method outperforms some recent related methods both in detection and classification. The highest accuracy of 99.994 ± 0.011 is achieved in case of detection, and 100% in case of pair-wise pathology classification.