Automatic Vocal Melody Extraction Via Quadratic Fluctuation Equation: A Comparative Analysis

Abstract

The extraction and recognition of vocal melodies from music data is an intricate but necessary step in digitized music technology. Efficient preprocessing methods are essential for precise musical signal evaluation and processing. Conventional techniques for automating this task include Convolutional Recurrent Neural Network-Conditional Random Field (CRNN-CRF), Non-Harmonic Adaptive NetworkGlobal Average Filtering (NHAN-GAF), and Frequency-Aware Multi-Objective Regression (FA-MOR), but they have constraints like lower accuracy and higher false alarm rates. These techniques frequently fail to sustain high precision in differentiating vocal melodies, resulting in suboptimal efficiency. Objectives: To tackle these drawbacks, the Quadratic Fluctuation Equation (QFE) is proposed as an innovative technique for automatically extracting vocal melodies. Methods: The QFE technique uses a Wiener filter and a penalized procedure to generate a dual-objective metric that efficiently manages phase discrepancies and reduces errors in inversion operations. This technique is especially good at compensating for problems like cyclical jumps in the frequency domain, which are common pitfalls in conventional techniques. Comprehensive computational experiments were carried out on a dataset of 373 ancient Chinese instrumental music pieces. The dataset, which included spectrograms from 17 various instruments, presented a solid foundation for assessing the effectiveness of the Quadratic Fluctuation Equation. Results: Experimental findings show that the Quadratic Fluctuation Equation surpasses previous approaches with an accuracy of 98%, which is an important advancement over modern techniques. The Quadratic Fluctuation Equation also performed well in terms of voice recall value, false alarm ratio, raw pitch precision, and raw chroma level. Conclusion: Overall, the Quadratic Fluctuation Equation technique is a robust solution for extracting and discriminating vocal melodies, with higher accuracy and dependability than previous methods. The findings highlight the capacity of the Quadratic Fluctuation Equation to advance the area of digitized music evaluation and signal processing.