Music main melody track identification method based on neural network

Abstract

The invention discloses a music main melody track identification method based on a neural network. The method comprises the following steps of extracting a plurality of features by utilizing an MIDI text to measure and describe each track in music, and automatically identifying and judging the main melody track of the music according to the basic features of the music by constructing a neural network model; the method comprises the following steps of (1) extracting the information of notes in music and calculating characteristic values of all audio tracks; (2) converting an audio track characteristic value into a two-dimensional neural network input matrix; (3) training a main melody audio track identification model based on a neural network; and (4) performing music main melody track identification by using the trained neural network. According to the method, important music features are extracted by using the MIDI file, and automatic identification is carried out after digital processing by using the neural network, so the main melody audio track can be accurately and efficiently detected and identified from the MIDI file; the music main melody track identification method based on the neural network is provided for research on automatic generation of music sequences.

Description

technical field [0001] The invention relates to a music main melody track identification method, in particular to a neural network-based music main melody track identification method. Background technique [0002] MIDI (Musical Instrument Digital Interface) file is a new music data recording format, which records music performance instruction sequences rather than actual sound information, which is more conducive to computer processing, and is automatically recognized for music features and automatically generated music sequences. Research provides great convenience. However, in the research on the automatic generation of music sequences, if the training set is multi-track music, the generated result is that the multi-track sounds are gathered in a single track, which not only deviates from the purpose of model learning, but also leads to automatically generated The quality of the music sequence is very poor; in addition, the extraction of the main melody has important appl...

AI Technical Summary

Problems solved by technology

However, in the research on the automatic generation of music sequences, if the training set is multi-track music, the generated result is that the multi-track sounds are gathered in a single track, which not only deviates from the purpose of model learning, but also leads to automatically generated The quality of the music sequence is very poor; in addition, the extraction of the main melody has important applications in humming retrieval, music transcription, music genre classification and singer identification, and is of great significance for improving the interactive experience of digital media and digital entertainment products

[0003] The existing main melody extraction algorithm mainly uses the sound analog data information contained in the music: the principle of a robust feature extraction algorithm for speech recognition is based on the main frequency information of the sub-band, and realizes the correlation between the main frequency information of the sub-band and the energy information of the sub-band. In combination, the sub-band peak position information in the spectrum is preserved in the feature parameters, and the anti-noise isolated word speech recognition system can be designed by using this algorithm; and the pitch sequence of the same sound source caused by the mutual interference of different sound sources in polyphonic music is different. Continuous, taking advantage of the continuity of pitch salience and the stability of higher harmonics, proposes methods for creating pitch contours based on pitch static likelihood functions and pitch salience dynamic likelihood functions, neither of which fully utilizes digitization Advantages of processing

And the task of detecting and identifying the main melody track from MIDI files still faces two challenges: (1) a piece of music is composed of singing voices and various musical instrument accompaniments, and the spectrum of different sound sources overlaps with each other, so it is difficult to distinguish a certain frequency component Attributed to a single sound source; (2) The research data source of digital music is different from the recording format of traditional music information, and the main melody information obtained from MIDI files cannot be processed according to traditional audio processing methods

Therefore, it is difficult for the prior art to detect and identify the main melody track accurately and quickly from the MIDI file

On the issue of building a classification model, although the rule-based classification method is easy to understand, it is not practical

Images