Musical Instrument Identification Using Deep Learning Approach.

Area of Science:

• Computer Science
• Music Information Retrieval
• Machine Learning

Background:

• Musical instrument identification is crucial for music analysis.
• Existing methods vary in input types and algorithms.
• A comprehensive review of related tasks and techniques is presented.

Purpose of the Study:

• To propose a novel approach for automatic musical instrument identification.
• To develop a system using individual convolutional neural networks (CNNs) for each instrument.
• To evaluate the efficiency and accuracy of the proposed model.

Main Methods:

• A review of musical instrument identification tasks, input types, algorithms, and metrics.
• Preparation and division of a dataset into training, validation, and evaluation subsets.
• Design and implementation of a neural network architecture with individual CNNs per instrument.

Main Results:

• The trained model demonstrated high efficiency across various instruments.
• Metric values ranged from 0.86 (guitar) to 0.99 (drums).
• Detailed precision, recall, and detection counts were reported for evaluation.

Conclusions:

• The proposed CNN-based approach is effective for automatic musical instrument identification.
• The model achieves high performance, indicating its potential for real-world applications.
• Further research can build upon this architecture for enhanced music information retrieval.