Framework for Groove Rating in Exercise-Enhancing Music Based on a CNN-TCN Architecture with Integrated Entropy...

Area of Science:

• Music Information Retrieval
• Computational Auditory Processing
• Machine Learning for Music

Background:

• Musical groove is vital for listener engagement but difficult to quantify due to complex acoustic features.
• Existing methods struggle to capture the intricate temporal dynamics and acoustic properties that define groove.

Purpose of the Study:

• To develop a novel computational framework for accurate groove prediction.
• To leverage deep learning architectures for analyzing audio features related to musical groove.

Main Methods:

• A hybrid Convolutional Neural Network (CNN) and Temporal Convolutional Network (TCN) framework was developed.
• Audio signals were transformed into Mel-spectrograms for feature extraction by CNNs and temporal analysis by TCNs.
• Entropy regularization and entropy pooling techniques were integrated to enhance model performance.

Main Results:

• The proposed CNN-TCN framework significantly outperformed benchmark methods in predicting musical groove.
• Entropy pooling and regularization were identified as critical components, with their absence reducing predictive accuracy (R²).
• The model demonstrated superior performance compared to standalone CNNs, LSTMs, and SVMs.

Conclusions:

• The developed CNN-TCN framework provides a robust method for automated musical groove assessment.
• This approach has potential applications in music education, therapy, and composition.
• Further research should focus on dataset expansion and model generalization for broader applicability.