基于神经网络与向量回归的旋律片段补全算法
Based on Neural Network and Vector Regression Melody Fragment Comple-tion Algorithm
张航 1李伟1
作者信息
- 1. 河南师范大学计算机与信息工程学院,河南新乡 453007
- 折叠
摘要
针对音乐编曲来说手动编曲整首旋律相对较难,文章提出一自动音乐旋律补全模型,基于SVR和多头LSTM.编曲者提供短旋律片段,转换为音高序列,通过正反向SVR和多头LSTM进行补全,提供灵活性.不同运算量的模型可同时生成多个补全结果,为编曲者提供更多选择和修改的可能性.相较以往正向补全研究,引入反向补全和多头LSTM扩展了补全方式,准确率相较于传统单向LSTM提升11％.
Abstract
When it comes to music composition,manually crafting the entire melody can be rela-tively challenging.This paper proposes an automated music melody completion model based on Support Vector Regression(SVR)and Multi-Head Long Short-Term Memory(LSTM).Com-posers provide short melody fragments,which are transformed into pitch sequences.The com-pletion process involves both forward and reverse SVR models,along with Multi-Head LSTM,introducing flexibility in the completion process.Models with varying computational complexi-ties can simultaneously generate multiple completion results,offering composers a broader range of choices and modification possibilities.In comparison to previous research focused solely on forward completion,the incorporation of reverse completion and Multi-Head LSTM expands the completion methodology,resulting in an 11％improvement in accuracy over traditional one-way LSTM approaches.
关键词
智能编曲/人工智能/神经网络/符号音乐Key words
intelligent composition/artificial intelligence/neural network/symbolic music引用本文复制引用
出版年
2024
NETL
NSTL
万方数据