Spectral-Temporal Receptive Field-Based Descriptors and Hierarchical Cascade Deep Belief Network for Guitar Playing...

Area of Science:

• Audio Signal Processing
• Music Information Retrieval
• Machine Learning

Background:

• Limited research exists on automatic classification of musical instrument playing techniques.
• Distinguishing subtle variations in guitar playing techniques (GPTs) presents a significant challenge.

Purpose of the Study:

• To develop an automatic system for classifying guitar playing techniques (GPTs).
• To introduce a novel framework for GPT classification leveraging advanced feature extraction and deep learning.

Main Methods:

• Feature extraction using spectral-temporal receptive fields (STRFs).
• Application of a supervised deep learning approach with a hierarchical cascade deep belief network (HCDBN).
• Evaluation using datasets comprising signal onsets, complete audio signals, and real-world audio environments.

Main Results:

• The proposed system achieved an F-score of 96.82% on complete audio signals.
• An improvement of approximately 11.47% in F-score was observed using signal onsets.
• The system demonstrated robust performance and high accuracy in real-world environments.

Conclusions:

• The developed HCDBN model and STRF feature extraction offer a robust and accurate solution for automatic GPT classification.
• The system's effectiveness in real-world scenarios highlights its practical applicability.
• This work advances the field of music information retrieval by addressing the nuanced classification of playing techniques.