Vocal Cords Diseases Detection By EEMD Based On Extension And Variable Cosine Window

Voice diseases rise due to environmental problems, social and occupational stress. Acoustic detection used for pathological voice recognition is still on the stage of theoretical research, and the study of characteristic parameters is still insufficient, voice acoustic feature optimization and the actual vocal physiology are lack of relations. As the most important vocal organ, vocal cords diseases are main cause of voice problems. The incidence of vocal cords polyp, vocal nodules and vocal cyst are the top three among the vocal cord diseases, and they are not easy to distinguish clearly by the acoustic detection, so the research of this three kinds of vocal cord diseases has become a hot spot.This paper analyses the vocal cords diseases voice. Limitation and deficiency will arise when traditional features describe influences of vocal cords diseases to voices signal, EEMD(Ensemble Empirical Mode Decomposition) based on Extension and Variable Cosine Window are proposed to improve the vocal cords diseases recognition rate. Specific studies are as follows:(1) The influences of vocal cords diseases to voice signal are introduced and reflections of traditional characteristics are also mentioned, and then correlation and difference analysis is carried out for feature selection. When researching the advantages and disadvantages of traditional EMD algorithm and improved algorithms(EEMD and multi-resolution EMD), variable cosine window function is introduced in detail, whose optimal parameter is chosen by comparative experiments of normal voice and vocal cord diseases voice. Recognition experiment of normal voice and vocal cord diseases voice and recognition experiment of normal speech、vocal cords polyp、vocal nodules and vocal cyst diseases show that marginal spectrum entropy and energy spectrum entropy extracted by this paper’s algorithm have higher recognition rate than traditional acoustic features.(2) This paper puts forward the new characteristic-sum of marginal spectrum on the basis of marginal spectrum, due to Hilbert-huang transform features cannot effectively characterize the characteristics of the vocal voice diseases. Recognition experiments also show that the proposed characteristic-sum of marginal spectrum can be subdivided of vocal cord diseases voice effectively.(3) The choicely traditional features and proposed features are used to recognition experiments to comprehensively describe influences of vocal cords diseases to vocal signal. The results show that the average recognition rate reaches 96.50% when sum of marginal spectrum is used for normal voices and vocal cord diseases, and average recognition rate of 75.05% for normal voices and vocal cord diseases(vocal cords polyp、vocal nodules and vocal cyst).