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

Abstract

Main melody extraction and multi-pitch estimation are two important research topics in the MIR field. In this article, the SVM algorithm is used to analyze and discuss music melody extraction and multi-pitch estimation. In the part of multi-fundamental frequency extraction, this article first filters the song signal with equal loudness and weakens the energy of the high-frequency and low-frequency parts of the song signal. Thereafter, the multi-resolution short-time Fourier transform suitable for processing song signals is introduced. In addition, in order to avoid the sharp jump of the estimated melody pitch in the same note duration range, this article proposes a main melody extraction method combining the SVM algorithm with dynamic programming. In this article, more features are used to distinguish the pitch contour of vocal fundamental frequency from that of the nonvocal fundamental frequency, which does not only depend on energy or a certain feature. The experimental results show that the lowest octave error of this method is 1.46. Meanwhile, the recall rate of the algorithm can reach about 95%. This method not only improves the recall rate of the fundamental frequency of the human voice but also improves the recall rate and pitch accuracy rate of the whole main melody extraction system.