Automatic Music Classification Model Based on Instantaneous Frequency and CNNs in High Noise Environment

Area of Science:

• Music Information Retrieval
• Signal Processing
• Machine Learning

Background:

• Automatic music classification is crucial for efficient music resource retrieval and applications.
• Existing methods require robust feature extraction and classification techniques.
• Deep learning (DL) offers powerful tools for analyzing complex data like music signals.

Purpose of the Study:

• To develop a DL-based model for music feature extraction and classification.
• To accurately estimate melody pitch sequences without relying on fundamental frequency.
• To leverage Convolutional Neural Networks (CNNs) for music feature classification.

Main Methods:

• Utilized instantaneous frequency and short-time Fourier transform for mixed music signal analysis.
• Employed a DL algorithm for pitch estimation based on peak-frequency pairs.
• Developed a CNN-based classifier using spectrograms as input for music feature classification.

Main Results:

• Achieved a music feature extraction accuracy of 95.66%.
• Achieved a music categorization accuracy of 94.18%.
• Demonstrated the model's effectiveness in extracting and classifying musical features.

Conclusions:

• The proposed DL-based approach is highly efficient for music feature extraction and classification.
• The method accurately estimates pitch sequences, enhancing music information retrieval capabilities.
• This technique shows significant promise for applications in the field of music information retrieval.