Deep Learning-Based Music Quality Analysis Model.

Area of Science:

• Computer Science
• Signal Processing
• Machine Learning

Background:

• Traditional music quality recognition relies on shallow learning with manual feature extraction.
• Deep learning offers advanced feature extraction capabilities, particularly with networks like PCANET.
• Integrating both approaches can potentially overcome individual limitations.

Purpose of the Study:

• To develop an efficient and effective deep music quality recognition model.
• To formulate a decision fusion method combining shallow and deep learning.
• To evaluate the proposed method's performance on established music databases.

Main Methods:

• A hybrid model is proposed, integrating shallow learning features with deep learning features extracted using PCANET.
• Spectrograms are used as input for the deep learning module.
• Support Vector Machine (SVM) is employed for music quality modeling, with decision fusion via differential voting.

Main Results:

• The proposed method significantly improves music quality recognition rates.
• Performance enhancements were observed on both a custom-compiled library and the Berlin database.
• The approach demonstrated clear advantages over competing methods.

Conclusions:

• The decision fusion of shallow and deep learning features provides a robust framework for music quality recognition.
• This hybrid approach offers superior performance compared to traditional methods.
• The study highlights the practical applicability of deep learning in music quality evaluation.