研究目的
Investigating the effectiveness of a novel feature extraction method combining wavelet packet decomposition and nonlinear feature extraction for classifying normal and pathological voices.
研究成果
The study demonstrates that vocal fold excitation signals extracted using wavelet packet transformation contain more useful information for pathological voice classification. The combination of Hurst parameter and second-order renyi entropy features achieves the highest classification accuracy of 99.21%, suggesting effective removal of vocal tract modulation information and highlighting the nonlinear features of voice.
研究不足
The study is limited by the specific subset of the MEEI database used and the focus on a particular method of feature extraction and classification. Potential areas for optimization include exploring other feature extraction methods and larger datasets.
1：Experimental Design and Method Selection:
The study employs wavelet packet decomposition to separate vocal fold excitation signals from speech signals, followed by nonlinear feature extraction and classification using Support Vector Machine (SVM).
2：Sample Selection and Data Sources:
Utilizes a subset of the Massachusetts Eye and Ear Infirmary (MEEI) database, including 53 normal and 173 pathological voice samples.
3：List of Experimental Equipment and Materials:
Uses wavelet packet transformation for signal decomposition and reconstruction, and SVM for classification.
4：Experimental Procedures and Operational Workflow:
Involves decomposing speech signals into 5 layers using wavelet packet, reconstructing high-frequency signals related to vocal fold, extracting nonlinear features, and classifying using SVM.
5：Data Analysis Methods:
Analyzes classification accuracy of different nonlinear features and their combinations.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
