CN-121999738-A - Music generation method and system integrating brain response signals and music semantic structures

CN121999738ACN 121999738 ACN121999738 ACN 121999738ACN-121999738-A

Abstract

The invention belongs to the technical field of music nerve stimulation, and relates to a music generation method and a system for fusing brain response signals and music semantic structures, wherein the method comprises the steps of acquiring nerve response signals of a subject in a music stimulation process, and extracting semantic structure characteristics of music; the method comprises the steps of aligning neural response signals and semantic structural features based on a cross attention mechanism, constructing a dynamic condition diffusion model based on the aligned neural response signals and semantic structural features to simulate a music-induced brain network reconstruction process, extracting characteristic parameters of high-response music fragments in the brain network reconstruction process, and generating personalized music tracks based on the characteristic parameters. The method has the beneficial effects that the limitation of subjective selection of the existing music tracks is broken through, a complete technical path from brain response identification to music generation is provided, and the method has wide application prospect and remarkable popularization value.

Inventors

WANG FEI
ZHU YUE
ZHANG HAIFENG

Assignees

南京医科大学

Dates

Publication Date: 20260508
Application Date: 20260203

Claims (10)

1. A music generation method integrating brain response signals and music semantic structures, comprising: Acquiring a neural response signal of a subject in a music stimulation process, and extracting semantic structural characteristics of music; aligning the neural response signals with the semantic structural features based on a cross-attention mechanism; Based on the aligned neural response signals and semantic structural features, constructing a dynamic condition diffusion model to simulate a music-induced brain network reconstruction process; Extracting characteristic parameters of a high-response music piece in the brain network reconstruction process; a personalized neural response music track is generated based on the feature parameters.
2. The method of claim 1, wherein obtaining a neural response signal of a subject during a musical stimulus comprises: Collecting cerebral cortex blood oxygen signals of a subject in a music stimulation process in real time by adopting a near infrared spectrum technology; preprocessing the cerebral cortex blood oxygen signals to obtain neural response signals.
3. The method for generating music by fusing brain response signals and music semantic structures as claimed in claim 2, the method is characterized by extracting semantic structural features of music, and comprises the following steps: extracting acoustic features of the musical material using a pre-trained music encoder; and acquiring the semantic structural features of the music by inputting the acoustic features of the musical material into the semantic analysis model.
4. A method of generating music that fuses brain response signals and musical semantic structures according to claim 3, wherein aligning neural response signals with semantic structure features based on a cross-attention mechanism comprises: constructing a multi-head cross attention network model; using the nerve response signal as a query vector, using the music semantic feature as a key vector, and generating a time-resolved modulation weight vector; And aligning the neural response signal with the semantic structural feature based on the time-resolved modulation weight vector to obtain the aligned neural response signal and the semantic structural feature.
5. The method for generating music by fusing brain response signals and music semantic structures as claimed in claim 4, wherein constructing a dynamic conditional diffusion model simulates a music-induced brain network reconstruction process, comprising: generating a target brain network structure through a conditional diffusion process based on the time-resolved modulation weight vector; optimizing the model to minimize reconstruction loss and structural similarity constraint, and completing the simulation of the music-induced brain network reconstruction process.
6. The music generation method of claim 5, wherein extracting characteristic parameters of a high-response music piece in a brain network reconstruction process comprises: Determining a time integral curve of the modulation weight in the brain network reconstruction process; dividing high/low response periods in the time integration curve based on the set response threshold; and extracting characteristic parameters of the music piece corresponding to the high response time period.
7. The method of claim 6, wherein generating a personalized neural response music track based on the feature parameters comprises: Performing acoustic feature inversion on the feature parameters to obtain common acoustic features among the high-response time periods, wherein the common acoustic features comprise rhythm patterns, spectrum energy distribution and tone; Analyzing the distribution rule of the common acoustic features; Based on the distribution rule of the common acoustic characteristics, optimizing the music generation parameters through reinforcement learning so as to generate personalized neural response music tracks according to the optimized music generation parameters.
8. A music generation system that fuses brain response signals with musical semantic structures, comprising: the nerve signal acquisition module is used for acquiring nerve response signals of the subject in the music stimulation process; The music feature extraction module is used for extracting semantic structural features of music; An alignment processing module for aligning the neural response signals with the semantic structural features based on a cross-attention mechanism; The network reconstruction module is used for constructing a dynamic condition diffusion model based on the aligned neural response signals and semantic structural features so as to simulate a music-induced brain network reconstruction process; the characteristic analysis module is used for extracting characteristic parameters of the high-response music piece in the brain network reconstruction process; and the music generation module is used for generating personalized neural response music tracks based on the characteristic parameters.
9. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements a music generation method of fusing brain response signals with music semantic structures as claimed in any one of claims 1 to 7.
10. A storage device comprising a storage medium storing a computer program and a processor, characterized in that the processor implements a music generation method of fusing brain response signals with music semantic structures according to any one of claims 1 to 7 when executing the computer program.

Description

Music generation method and system integrating brain response signals and music semantic structures Technical Field The invention relates to the technical field of biomedical engineering, in particular to a music generation method and system integrating brain response signals and music semantic structures. Background In the biomedical engineering field, music is used as an adjustable complex acoustic stimulus, and has wide application prospect in the research of nervous system modulation and man-machine interaction. However, the existing technical solution based on the music nerve stimulation still has significant defects in the aspects of music track generation and parameterization design: 1. The music track selection is modeled according to the lack of a neural mechanism, and the intelligent degree is insufficient, so that the existing music recommendation system depends on subjective experience or simple preference matching, and an intelligent recommendation model based on the neural mechanism cannot be built. For example, experience driven recommendation, where traditional music track selection mainly employs static song list recommendation (e.g., relaxing music collection, classical music stimulation track library), lacks dynamic recommendation capabilities based on individual brain loop characteristics. 2. And the neural response matching is missing, namely the recommendation algorithm does not integrate specific neural response data of the user on the music structure (rhythm, harmony and the like), so that the recommendation result has low matching degree with the individual neural regulation requirement. Dynamic neural response is disjointed with real-time recommendations. 3. The closed loop optimization system of music generation and neural response optimization is not realized in the prior art, the closed loop optimization system of music parameters-brain state-recommendation strategy is not realized, the existing music tracks mostly adopt a passive play mode, and a common music relaxation chair or play list only provides unidirectional stimulation and cannot adjust the music parameters according to real-time brain connection changes. 4. The interdisciplinary technology is not deeply integrated into the recommendation algorithm, and the rupture still exists between the neural mechanism of music stimulation and the artificial intelligence generation technology. Brain science data is not fully transformed-although near infrared spectroscopy techniques have demonstrated that music can regulate forehead lobe function, such findings have not been systematically applied to music generation algorithms. The artificial intelligence generation technology lacks clinical verification in the use of music, and the current AI composing technology (such as MusicGAN) pays attention to artistry more, and does not take curative effect as an optimization target of a recommendation algorithm. In summary, the related music track selection has core defects of strong subjectivity, lack of neural response matching, insufficient closed-loop regulation, weak interdisciplinary integration and the like, and the limitations severely restrict the development of the music neural stimulation to the intelligent and personalized recommendation direction. Disclosure of Invention Technical problem to be solved In view of the above-mentioned shortcomings and disadvantages of the prior art, the present invention provides a method and a system for generating music by fusing brain response signals and music semantic structures, which solve the technical problems of strong subjectivity, lack of neural response matching, insufficient closed-loop regulation and weak interdisciplinary integration. Technical proposal In order to achieve the above purpose, the main technical scheme adopted by the invention comprises the following steps: In a first aspect, the present invention provides a music generation method of fusing brain response signals with music semantic structures, comprising: Acquiring a neural response signal of a subject in a music stimulation process, and extracting semantic structural characteristics of music; aligning the neural response signals with the semantic structural features based on a cross-attention mechanism; Based on the aligned neural response signals and semantic structural features, constructing a dynamic condition diffusion model to simulate a music-induced brain network reconstruction process; Extracting characteristic parameters of a high-response music piece in the brain network reconstruction process; a personalized neural response music track is generated based on the feature parameters. Optionally, acquiring a neural response signal of the subject during the musical stimulus, comprising: Collecting cerebral cortex blood oxygen signals of a subject in a music stimulation process in real time by adopting a near infrared spectrum technology; preprocessing the cerebral cortex blood oxygen signals to obtain neu