Extraction of Music Main Melody and Multi-Pitch Estimation Method Based on Support Vector Machine in Big Data...

Area of Science:

• Music Information Retrieval (MIR)
• Signal Processing
• Machine Learning

Background:

• Main melody extraction and multi-pitch estimation are crucial in MIR.
• Existing methods face challenges in accurately identifying fundamental frequencies, especially in complex audio signals.

Purpose of the Study:

• To develop an improved method for main melody extraction and multi-pitch estimation.
• To enhance the accuracy and recall rate of fundamental frequency detection, particularly for vocal components.

Main Methods:

• Utilized the Support Vector Machine (SVM) algorithm for analysis.
• Applied equal loudness filtering and multi-resolution short-time Fourier transform for signal processing.
• Combined SVM with dynamic programming to refine melody pitch estimation and avoid abrupt changes.
• Incorporated diverse features to differentiate vocal from non-vocal fundamental frequencies.

Main Results:

• Achieved the lowest octave error of 1.46.
• Reached a recall rate of approximately 95% for the algorithm.
• Demonstrated significant improvements in the recall rate of human voice fundamental frequency.
• Enhanced the overall recall rate and pitch accuracy of the main melody extraction system.

Conclusions:

• The proposed SVM and dynamic programming combined method effectively improves main melody extraction and multi-pitch estimation.
• The method shows superior performance in identifying vocal fundamental frequencies with high accuracy and recall.
• This research contributes to advancing MIR technologies for more precise music analysis.