Automatic generation method for civil song melody based on self-attention mechanism

Abstract

The invention discloses a method for automatically generating a civil song melody based on a self-attention mechanism, and relates to the technical field of intelligent melody generation, and the method comprises the steps: firstly carrying out a series of preprocessing on a civil song data set, and converting the data into a time sequence suitable for network training; then sequentially utilizing CNN to extract melody local features, utilizing a self-attention mechanism to capture key features, utilizing BiGRU to extract global features and the like, constructing a CNN-BiGRU neural network fused with the self-attention mechanism, and training the constructed neural network by using a data set obtained by preprocessing; and finally, utilizing the trained network to carry out prediction generation of the civilian song melody. According to the method, various evaluation indexes are improved, the generated music melody is natural and graceful, and the method has the capability of generating novel music paragraphs.

Description

technical field [0001] The invention relates to the technical field of intelligent melody generation, in particular to an automatic generation method of folk song melody based on a self-attention mechanism. Background technique [0002] Music has an important impact on human life. With the continuous development of computer technology and deep learning, the automatic creation of music has become one of the current research hotspots. To this end, scholars have proposed many related research attempts, most of which focus on automatic composition; in recent years, with the continuous development of deep learning technology, researchers have also used related technologies and networks to conduct music generation research. But overall, due to the complexity of the research objects, automatic music generation is still a challenging task. [0003] The existing deep learning neural network composition mainly includes recurrent neural network (RNN, Recurrent Neural Network) and long...

AI Technical Summary

Problems solved by technology

However, the music generated based on a single type of neural network lacks continuity, the theme melody is not prominent, the beginning and the end are unnatural, the generation efficiency is low, and the audibility is relatively poor.

At the same time, most of the existing research in the field of music generation focuses on the generation of classical piano music and jazz, and there are relatively few studies on the generation of Chinese folk song melodies.

Images