Digital music synthesis method and device for pulsar signal control

Abstract

The invention relates to the field of digital signal processing, and discloses a digital music synthesis method and device for pulsar signal control, and the method comprises the steps: building an automatic encoder model, training the automatic encoder model, obtaining the trained automatic encoder model, and obtaining a pulsar signal, performing feature extraction on the signals of the plurality of pulsar by using a fundamental frequency encoder and a loudness encoder to obtain the feature representation of the pulsar; and performing a digital music synthesis task of pulsar signal control on the audio signal of the music according to an automatic encoder model and the pulsar characteristic representation. According to the method, a deep convolutional neural network model is used for extracting fundamental frequency features, an A-weighted power spectrum is used for obtaining loudness features, a least square method is combined for solving a multivariate linear equation set, a control weight signal of pulsar signals is obtained, the contribution of each pulsar in the music synthesis process is displayed in real time, the model parameter scale is effectively reduced, and the music synthesis efficiency is high.

Description

technical field [0001] The invention relates to the technical field of digital signal processing, in particular to a digital music synthesis method and equipment for pulsar signal control. Background technique [0002] In the prior art, the music synthesis model based on deep learning generally directly generates the time-domain waveform or spectrogram distribution of music. Although it is sufficient to represent any music signal, these methods do not take advantage of the physical principles of sound generation, propagation and perception, and the model parameters Huge scale, low efficiency. [0003] For example, the national patent publication CN107871492A discloses a "music synthesis method and system", including obtaining the sound information to be synthesized, and obtaining the corresponding linear prediction coefficient according to the sound information to be synthesized; obtaining the linear prediction coefficient of the sound information to be synthesized according...

AI Technical Summary

Problems solved by technology

The music synthesis method provided by the invention directly generates the time-domain waveform and spectrum data of music. Although it is sufficient to represent the music signal, the invention does not use the physical principles of sound generation, propagation and perception. The model parameters are huge and the music synthesis efficiency is low.

Images