Quality of the human voice can be affected by anomalies of the larynx due to the physical, Nerve-muscle or only nervous origins. Video Stroboscope and vocal folds movement display systems are key tools which often used to detect Laryngeal anomalies. These methods are invasive, time consuming and expensive, so researchers are trying to find non-invasive methods that lead to the final answers faster than invasive methods and contain tolerable condition for patients. Many interests are directed to the application of speech processing techniques in relevant works. In these works, researchers were used different processing methods in medical engineering to detect anomalies. Recently, variety of researches presented to detect anomalies from the audio signals of individuals based on the features that extracted from audio signals. These methods have been conducted to separate patient audio from non-patient once. These researches do not work properly when an anomaly is among several anomalies and achieve bad error rate. In this paper, we aim to propose a new method of automatic Anomalies detection which performs based on a new mechanism of feature extraction and a PSO classifier. In the proposed work, Feature extraction is done in three ways, the first depending on MFCC features and the second depending on Jitter and Shimmer features and the third by combining MFCC and Jitter and Shimmer. Meanwhile, achieved features are used along with PSO algorithm to analysis and classify anomalies based on several classes. Also, we used four groups of anomalies and a class of normal voice as benchmark data sets and evaluated and compared the proposed method with different feature extraction strategy. Our simulations results confirm the superior performance of the proposed method, especially when the features are extracted based on combination of MFCC and Jitter Shimmer. The result from the combination is 80% and using MFCC alone is 66% and using Shimmer and Jitter is 43%.
