Emotion Recognition of Violin Playing Based on Big Data Analysis Technologies.

Area of Science:

• Music Information Retrieval
• Artificial Intelligence
• Signal Processing

Background:

• The proliferation of digital multimedia and violin performances necessitates efficient organization and retrieval systems.
• Classifying music by emotional properties is a common but challenging task.
• Deep learning models, particularly Recurrent Neural Networks (RNNs) like Long Short-Term Memory (LSTM), offer powerful capabilities for analyzing time-series data.

Purpose of the Study:

• To propose and evaluate an emotion recognition method for dynamic violin performances using LSTM.
• To leverage acoustic features and the Hevner emotion classification model for accurate music emotion analysis.
• To enhance the organization and retrieval of violin musical works based on their emotional content.

Main Methods:

• Utilized the Long Short-Term Memory (LSTM) deep learning model, a variant of Recurrent Neural Networks (RNNs).
• Selected relevant acoustic features from violin performance audio signals.
• Employed the Hevner emotion classification model as a basis for emotional categorization.
• Conducted data labeling, feature selection, and model testing on actual violin music datasets.

Main Results:

• Achieved a prediction accuracy of 83% for emotion recognition in violin performances, surpassing existing methods.
• Demonstrated a significant reduction in training time compared to conventional approaches.
• Evaluated accuracy and iteration times across different emotional categories of violin music.
• Confirmed the method's robustness against variations in genre, timbre, and noise.

Conclusions:

• The proposed LSTM-based method offers a superior approach to emotion recognition in dynamic violin performances.
• The technique significantly improves prediction accuracy and reduces computational overhead.
• The robustness of the method makes it applicable to diverse real-world scenarios in music information retrieval.