Banca de DEFESA: WINNIE DE LIMA TORRES

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : WINNIE DE LIMA TORRES
DATA : 30/01/2018
HORA: 10:00
LOCAL: Laboratório de Automação, Controle e Instrumentação (LACI)
TÍTULO:

Detection of Vocal Deviations Using Auto Regressive Models and the KNN Algorithm

PALAVRAS-CHAVES:

Detection of Vocal Deviations, Auto Regressive Models, k Nearest Neighbor

PÁGINAS: 65
GRANDE ÁREA: Engenharias
ÁREA: Engenharia Elétrica
SUBÁREA: Eletrônica Industrial, Sistemas e Controles Eletrônicos
ESPECIALIDADE: Controle de Processos Eletrônicos, Retroalimentação
RESUMO:

Some fields of science propose to study disorders in the vocal tract from analyzes on patterns of vibration of the voice. In general, the importance of these researches is to identify, in a more specific stage, diseases of greater or lesser severity, to be healed with vocal therapy or that require more attention, even generating the need for surgical procedures for its control. Although there are already indications in the literature that digital signal processing allows non-invasive diagnosis of laryngeal pathologies, such as vocal diseases that cause edema, nodule and paralysis, there is no definition of the most indicated method and the most characteristic features or parameters to detect the presence of vocal deviations. Thus, in this work an algorithm is proposed for the detection of vocal deviations through the analysis of speech signals. To perform this work, we used constant data in the database Disordered Voice Database, developed by the Massachusetts Eye and Ear Infirmary (MEEI), due to its use in research in the acoustic area of voice. We used 166 signals contained in this database, with signs of healthy voices and pathological voices affected by edema, nodule and paralysis in the vocal folds. From the voice signals, Auto Regressive models (AR and ARMA) were generated to represent these signals and, using the parameters of the obtained models, the KNN algorithm was used to classify the analyzed signals. In order to analyze the efficiency of the algorithm proposed in this study, the results obtained from this algorithm were compared with a detection method considering only Euclidean distance between the signals. The results show that the method proposed in this work presents a good result, generating a rating rate above 71% (greater than 31% from the use of Euclidean distance). In addition, the method used is easy to implement and can be used in simpler hardware. Therefore, this research has the potential to generate an inexpensive and affordable classifier for large-scale use by health professionals as a non-invasive pre-analysis alternative for the detection of otorhinolaryngological pathologies that affect the voice.

MEMBROS DA BANCA:
Presidente - 347565 - ALDAYR DANTAS DE ARAUJO
Interno - 2579664 - ALLAN DE MEDEIROS MARTINS
Externo à Instituição - ADEMAR GONÇALVES DA COSTA JÚNIOR - IFPB