A mood symbol music generation method and device and a storage medium

By using the CP-Transformer deep learning model and compound word representation, music attributes are discretized, solving the problems of high model training overhead and inflexible control signal representation in existing systems, and achieving efficient generation of emotional symbol music.

CN117953838BActive Publication Date: 2026-06-09SOUTH CHINA UNIV OF TECH

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SOUTH CHINA UNIV OF TECH
Filing Date
2024-01-30
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing emotion symbol music generation systems require determining the control signal set and model design during model pre-training, and retraining is required during adjustment, increasing time and space overhead. Existing systems use one-dimensional representations, which are difficult to handle sequential control signals, and the control signals are usually binary, making the representation inflexible.

Method used

The CP-Transformer deep learning model is used to discretize music attributes through an attribute preprocessing module, encode control signals using compound word representation, and combine music generation and control modules to achieve the generation of emotion symbol music, allowing fine-tuning based on the weights of a unified pre-trained model.

Benefits of technology

It enables flexible adjustment of model settings, reduces the time and space overhead of model training, improves the flexibility and emotional intensity of emotional symbol music generation, and retains musicality while adapting to personalized control needs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117953838B_ABST
    Figure CN117953838B_ABST
Patent Text Reader

Abstract

The application discloses a mood symbolic music generation method and device and a storage medium, and belongs to the technical field of data processing. The method comprises the following steps: obtaining input information, generating mood labels required for music according to the input information; mapping the mood labels to obtain a music attribute set based on a preset mapping relationship between moods and music attributes; encoding the mood labels and the music attribute set to obtain a control composite event sequence; generating a splicing vector of fusion control information according to the control composite event sequence; predicting symbolic music composite events one by one in an autoregressive manner according to the splicing vector of fusion control information to obtain a music composite event sequence; and decoding the music composite event sequence to obtain symbolic music. The application separates the encoding sequences of music and control signal sets and corresponding processing modules, flexibly adjusts model settings related to music control, and only needs to perform model fine tuning with small time and space overhead, so that the model has the ability to generate mood symbolic music.
Need to check novelty before this filing date? Find Prior Art