five

古筝真实演奏数据集 Guzheng Tech99 Dataset

收藏
魔搭社区2026-05-08 更新2024-05-15 收录
下载链接:
https://modelscope.cn/datasets/ccmusic-database/Guzheng_Tech99
下载链接
链接失效反馈
官方服务:
资源简介:
大部分的古筝独奏存在“重叠技法”(多个使用不同技法的音符同时弹奏)或者“混合技法”(一个音符使用了多种技法,例如:“滑摇”就是“滑音”和“摇指”的结合)的情况,而大部分现有的对于乐器演奏技法检测的研究只在不存在重叠演奏技法的乐器上进行,所以该数据集的提出填补了该领域研究的一块空白,增加乐器演奏技法检测领域的研究范围和多样性。该数据集包含了99首古筝独奏曲目,由两位古筝演奏家在专业录音棚录制而成,涵盖了古筝音乐的大部分类型。数据集中的音频记录总共长9064.6秒。我们按照古筝简谱对这些古筝曲中存在的7种演奏技法进行标注(颤音、点音、上滑音、下滑音、拨弦、刮奏、摇指),我们对录音中的每个音符都标注出了起始时间、结束时间、音高和演奏技法。最终,该数据集总共包含了63352个标注。该数据集在训练集、验证集和测试集中分别分配了79首、10首和10首歌曲。

Most solo guzheng performances feature either overlapping playing techniques (multiple notes with distinct techniques played simultaneously) or combined playing techniques (one note employs multiple techniques; for example, "sliding tremolo" combines glissando and finger tremolo). However, most existing studies on musical instrument playing technique detection only focus on instruments that do not have overlapping playing techniques. Thus, the release of this dataset fills a critical research gap in this field and expands the scope and diversity of research in the area of musical instrument playing technique detection. This dataset contains 99 pieces of guzheng solo music, recorded by two professional guzheng performers in a professional recording studio, covering most genres of guzheng music. The total duration of the audio recordings in this dataset is 9064.6 seconds. We annotated 7 types of playing techniques present in these guzheng pieces based on guzheng numbered musical notation: vibrato, staccato pluck, upward glissando, downward glissando, pluck, glissando sweep, and finger tremolo. For each note in the recordings, we annotated its start time, end time, pitch, and corresponding playing technique. In total, this dataset contains 63,352 annotations. This dataset is split into training, validation, and test sets with 79, 10, and 10 pieces respectively.
提供机构:
maas
创建时间:
2024-07-10
搜集汇总
数据集介绍
main_image_url
背景与挑战
背景概述
该数据集包含99首由专业音乐家录制的古筝独奏作品,每首作品均进行了音符级别的详细注释,涵盖颤音、拨弦等七种演奏技巧,共产生63,352个标签。经过集成处理,数据集优化为包含音频、梅尔频谱图和标签的结构,并提供了用于训练和评估的子集,但存在数据不平衡问题,例如拨弦技巧占比显著偏高。
以上内容由遇见数据集搜集并总结生成
二维码
社区交流群
二维码
科研交流群
商业服务