Music generation method based on variable neighborhood search and masking effect

Abstract

The invention relates to the field of music generation, in particular to a music generation method based on variable neighborhood search and a masking effect. The problem that as for an existing method, music generation cannot be conducted according to music clips obtained through the sudden inspiration or extemporaneous play of a composer, and the problem that generated music may have cluttered notes are solved. Time domain training data are subjected to melody extraction based on the masking effect, mapped to a note domain and visualized; a to-be-trained music score is input into a model toobtain a parameter matrix, and the parameter matrix can be mixed with existing training model parameters; at the generation stage, based on a variable neighborhood search algorithm, the music score isiterated repeatedly before meeting the threshold, and grading is conducted; and finally, subsequent sections or accompaniments and drumbeats can be selected to be generated. The music score can be manually corrected and annotated nearly every step, and the threshold of composing can be lowered while repeatable work is reduced; and the music generation method is mainly used for music generation.

Description

technical field [0001] The present invention relates to the field of music generation. Background technique [0002] Music plays a unique role in daily life. People use music to release stress, express emotions, seek resonance, or enjoy the beautiful experience brought by music. With the continuous improvement of living standards, ordinary people can sing or play beautiful melodies with a little practice, but the threshold for composing is obviously higher in comparison. Compiling a piece of music is often divided into parts such as melody, style, speed, tonality and structure, harmony and instruments, and the various parts of the piece are not completely orthogonal, but interrelated. For example, the melody of a song often determines the Its style, and the style affects the speed to a certain extent; phrases are often repetitions or variations of bars, etc. Therefore, in the past, composing music belonged to the work of a few talented people. At the same time, composing a...

AI Technical Summary

Problems solved by technology

However, the current music generation methods (such as music generation methods based on long-short-term memory networks, recurrent neural networks, or convolutional neural networks) cannot perform music generation according to the composer's sudden inspiration or music fragments obtained by improvisation, and the generated The music is completely automatically generated, and it does not fully comply with the composer's wishes, and even differs greatly from the composer's wishes

Moreover, the music generated by the existing method may appear to have a compact melody on the surface, but in fact it may be music with chaotic notes.

Images