一种变邻域搜索与人耳掩蔽音乐生成方法  

A music generation method based on variable neighborhood search and masking effect

作　　者：梅林 肖兆雄 沙学军[1] MEI Lin;XIAO Zhaoxiong;SHA Xuejun(School of Electronics and Information Engineering,Harbin Institute of Technology,Harbin 150080,China)

机构地区：[1]哈尔滨工业大学电子与信息工程学院,哈尔滨150080

出　　处：《哈尔滨工业大学学报》2020年第5期1-8,共8页Journal of Harbin Institute of Technology

摘　　要：本文提出了一种基于变邻域搜索与人耳掩蔽效应的音乐生成方法,以克服传统的作曲方法门槛过高、可重复性工作较多等缺点,并且相较于现有音乐生成模型具有操作直观、能够挖掘乐谱长时间内的连贯性等优点.训练阶段的输入选择易获得的时域波形文件,通过基于音高显著度的频率提取和基于掩蔽效应的时域分块获得乐谱,经过预先设定的音乐参数进行训练.生成阶段采取变邻域搜索与人工操作相结合的方式,通过对乐谱中音符进行增删、转调等操作按小节生成具有艺术性的乐谱,并根据操作者的偏好对乐谱进行人工修改与锁定.随后,以4小节为单位一次生成音乐片段,并组合成持续时间较长的一段音乐片段.最后在不同时间尺度完善乐谱,输出时域波形文件.整个流程可视化和人机交互的思想贯穿其中,能够充分调动音乐爱好者的积极性.同时本文提出了一种参数化乐音模拟的方法用于修饰乐谱或生成全新的音色,可根据乐器的分类方法或人的发声过程,通过声源激励与共振腔模拟两个部分参数化模拟乐器与人声.A music generation method based on vaoabOe neighborhood seerch and masking effect was proposed tu overccme the shortcomings of traditional composing methods,such as high threshold and repeatable work.Compared with the existing music generation model,the proposed method is intuitiva tu operate and con excovate the consistency of music scores ovec a long period of time.In the training staae,the time domain wavaform files were chosen as the input,which wera esy to acquire.The musio scoa was obtained by frequency extraction based on pitch signinconco and time domain block based on masking effect,and tie music parametera were pre-set foe training.In the generation stage,the method of combining variable neighborhood search witi manual operation was adopted.By adding,dceting,and changing the notes in tee musio score,the artistio musio score was generated according te the section,and the music score was manudy modified and locked according te the preferenco of tee operatoe.Then,tee music segments were generated in foue sections at a time and combined inte a long-lasting music segment.Findy,tee music score was perfected at dnferent time sccles,and the time domain waveform file was output.The inea of visualization and humaniomputer interaction teroughout the whole process could fully mobilize the enteusiasm of music enthusiasts.Besides,a parametec musiccl instruments simulation mettod is proposed te modiff the music score or generate a new timbre,in which two parts of parametec sinulation of musiccl instruments and human vaico con be ccrried out through sound sourco excitation and resonant cavity sinulation according te the classificytion of musiccl instruments or tee process of human vaico production.

关 键 词：音乐生成 变邻域搜索 掩蔽效应 人机交互 音高显著度 乐音模拟 

分 类 号：TN912[电子电信—通信与信息系统]