融合乐理规则与AC算法的AI多音轨音乐生成研究  

Research on AI multi-track music generation by integrating music theory rules and AC algorithm

作　　者：张戈 ZHANG Ge(Shaanxi Polytechnic Institute,XianYang Shanxi 712000,China)

机构地区：[1]陕西工业职业技术学院,陕西咸阳712000

出　　处：《自动化与仪器仪表》2024年第10期178-181,186,共5页Automation & Instrumentation

基　　金：陕西省职业技术教育学会2022年度一般研究课题《以红色经典艺术为载体培育学生技能报国理想的探索与实践》(2022SZX167)。

摘　　要：由于人工智能(Artificial Intelligence,AI)受限于数据的输入,在生成的音乐中缺乏部分旋律和节奏的完整性。研究为了解决这个问题,提出了一种基于旋律与节奏生成,并结合演员-评论家(Actor-Critic,AC)算法的多音轨音乐生成模型。研究首先利用AC算法设计旋律和节奏模型,然后在此基础上利用强化学习进行改进得到多音轨音乐生成模型。结果表明生成模型的数据处理准确率和误差率分别为93.28%、0.16%,生成模型的训练损失值在迭代至98次时平均损失值为0.18、音轨匹配度为0.83,所有结果指标均优于对比方法。这说明该生成模型具有较高的鲁棒性和准确性,能够为多音轨音乐的生成提供新的路径。Due to limitations in data input,artificial intelligence(AI)lacks the integrity of some melodies and rhythms in the generated music.To address this issue,a multi track music generation model based on melody and rhythm generation,combined with the Actor Critic(AC)algorithm,was proposed.The study first uses the AC algorithm to design melody and rhythm models,and then improves them using reinforcement learning to obtain a multi track music generation model.The results show that the data processing accuracy and error rate of the generated model are 93.28%and 0.16%,respectively.The average training loss value of the generated model after 98 iterations is 0.18,and the track matching degree is 0.83.All result indicators are better than the comparison method.This indicates that the generation model has high robustness and accuracy,and can provide a new path for the generation of multi track music.

关 键 词：AC算法 多音轨音乐 音乐旋律 音乐节奏 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]