Audio Commons Ground Truth Data for deliverables D4.4, D4.10 and D4.12

This dataset contains the ground truth data used to evaluate the musical <strong>pitch</strong>, <strong>tempo</strong> and <strong>key </strong>estimation algorithms developed during the AudioCommons H2020 EU project and which are part of the Audio Commons Audio Extractor tool. It also includes ground truth information for the <strong>single-event<em>ness</em> </strong>audio descriptor also developed for the same tool. This ground truth data has been used to generate the following documents: <strong>Deliverable D4.4</strong>: Evaluation report on the first prototype tool for the automatic semantic description of music samples <strong>Deliverable D4.10</strong>: Evaluation report on the second prototype tool for the automatic semantic description of music samples <strong>Deliverable D4.12</strong>: Release of tool for the automatic semantic description of music samples All these documents are available in the materials section of the AudioCommons website. All ground truth data in this repository is provided in the form of CSV files. Each CSV file corresponds to one of the individual datasets used in one or more evaluation tasks of the aforementioned deliverables. This repository <strong>does not include the audio files</strong> of each individual dataset, but includes references to the audio files. The following paragraphs describe the structure of the CSV files and give some notes about how to obtain the audio files in case these would be needed. <br> <strong>Structure of the CSV files</strong> All CSV files in this repository (with the sole exception of <em>SINGLE EVENT - Ground Truth.csv</em>) feature the following 5 columns: <strong>Audio reference</strong>: reference to the corresponding audio file. This will either be a string withe the <strong>filename</strong>, or the <strong>Freesound ID </strong>(for one dataset based on Freesound content). See below for details about how to obtain those files. <strong>Audio reference type</strong>: will be one of <em>Filename</em> or <em>Freesound ID</em>, and specifies how the previous column should be interpreted. <strong>Key annotation</strong>: tonality information as a string with the form "RootNote minor/major". Audio files with no ground truth annotation for tonality are left blank. Ground truth annotations are parsed from the original data source as described in the text of deliverables D4.4 and D4.10. <strong>Tempo annotation</strong>: tempo information as an integer representing beats per minute. Audio files with no ground truth annotation for tempo are left blank. Ground truth annotations are parsed from the original data source as described in the text of deliverables D4.4 and D4.10. Note that integer values are used here because we only have tempo annotations for <em>music loops</em> which typically only feature integer tempo values. <strong>Pitch annotation</strong>: pitch information as an integer representing the MIDI note number corresponding to annotated pitch's frequency. Audio files with no ground truth pitch for tempo are left blank. Ground truth annotations are parsed from the original data source as described in the text of deliverables D4.4 and D4.10. The remaining CSV file, <em>SINGLE EVENT - Ground Truth.csv</em>, has only the following 2 columns: <strong>Freesound ID</strong>: sound ID used in Freesound to identify the audio clip. <strong>Single Event: </strong>boolean indicating whether the corresponding sound is considered to be a single event or not. Single event annotations were collected by the authors of the deliverables as described in deliverable D4.10. <strong>How to get the audio data</strong> In this section we provide some notes about how to obtain the audio files corresponding to the ground truth annotations provided here. Note that due to licensing restrictions we are not allowed to re-distribute the audio data corresponding to most of these ground truth annotations. <strong>Apple Loops (APPL)</strong>: This dataset includes some of the music loops included in Apple's music software such as Logic or GarageBand. Access to these loops requires owning a license for the software. Detailed instructions about how to set up this dataset are provided here. <strong>Carlos Vaquero Instruments Dataset (CVAQ)</strong>: This dataset includes single instrument recordings carried out by Carlos Vaquero as part of this master thesis. Sounds are available as Freesound packs and can be downloaded at this page: https://freesound.org/people/Carlos_Vaquero/packs <strong>Freesound Loops 4k (FSL4)</strong>: This dataset set includes a selection of music loops taken from Freesound. Detailed instructions about how to set up this dataset are provided here. <strong>Giant Steps Key Dataset (GSKY)</strong>: This dataset includes a selection of previews from Beatport annotated by key. Audio and original annotations available here. <strong>Good-sounds Dataset (GSND)</strong>: This dataset contains monophonic recordings of instrument samples. Full description, original annotations and audio are available here. <strong>University of IOWA Musical Instrument Samples (IOWA)</strong>: This dataset was created by the Electronic Music Studios of the University of IOWA and contains recordings of instrument samples. The dataset is available upon request by visiting this website. <strong>Mixcraft Loops (MIXL)</strong>: This dataset includes some of the music loops included in Acoustica's Mixcraft music software. Access to these loops requires owning a license for the software. Detailed instructions about how to set up this dataset are provided here. <strong>NSynth Dataset Test and Validation sets (NSYT and NSYV)</strong>: NSynth is a large-scale and high-quality dataset of annotated musical notes built with synthesized sounds by Google's Magenta team. Full dataset description including original annotations and audio files is available here. <strong>Philarmonia Orchestra Sound Samples Dataset (PHIL)</strong>: This includes thousands of free, downloadable sound samples specially recorded by Philharmonia Orchestra players. Audio files are freely downloadable from the philarmonia orchestra website. <strong>Freesound Single Events Dataset (SINGLE EVENT)</strong>: This includes a selection of Freesound audio clips representing audio signals containing either a single audio <em>event</em> or multiple ones. Original audio files can be retrieved by downloading individual audio clips from Freesound using the ID identifier provided in the CSV file. A similar procedure to that described here could be followed.本数据集包含用于评估**音高（pitch）**、**速度（tempo）**与**调式（key）**估计算法的基准真值（ground truth）数据，相关算法由欧盟H2020计划下的AudioCommons项目研发，且为Audio Commons音频提取器工具的组成部分。本数据集同时包含为该工具研发的**单事件性（single-event-ness）**音频描述符的基准真值信息。 本基准真值数据集已用于生成以下交付文档： **交付文档D4.4**：音乐样本自动语义描述首个原型工具评估报告 **交付文档D4.10**：音乐样本自动语义描述第二版原型工具评估报告 **交付文档D4.12**：音乐样本自动语义描述工具发布 上述所有文档均可在AudioCommons官网的资料板块获取。 本仓库内的所有基准真值数据均以CSV（逗号分隔值）文件形式提供。每个CSV文件对应上述交付文档的一项或多项评估任务所使用的独立数据集。本仓库**不包含各独立数据集的音频文件**，但提供音频文件的引用信息。下文将阐述CSV文件的结构，并提供所需音频文件的获取指南。 ### CSV文件结构 本仓库内除《SINGLE EVENT - Ground Truth.csv》外的所有CSV文件，均包含以下5列： 1. **音频引用（Audio reference）**：对应音频文件的引用标识，可为由文件名组成的字符串，或基于Freesound内容的数据集所用的**Freesound标识符（Freesound ID）**。获取这些文件的详细说明见下文。 2. **音频引用类型（Audio reference type）**：取值为“Filename（文件名）”或“Freesound ID（Freesound标识符）”，用于指定前一列内容的解析方式。 3. **调式标注（Key annotation）**：调性信息，格式为字符串“根音 小调/大调”。无调式基准真值标注的音频文件对应字段留空。基准真值标注源自原始数据源，具体解析方式详见交付文档D4.4与D4.10的正文内容。 4. **速度标注（Tempo annotation）**：速度信息，以代表每分钟节拍数（beats per minute，BPM）的整数形式呈现。无速度基准真值标注的音频文件对应字段留空。基准真值标注源自原始数据源，具体解析方式详见交付文档D4.4与D4.10的正文内容。需注意，此处仅使用整数取值，因为本数据集仅包含音乐循环（music loops）的速度标注，而此类循环通常仅具有整数节拍速度。 5. **音高标注（Pitch annotation）**：音高信息，以代表对应音高频率的MIDI音符编号的整数形式呈现。无音高基准真值标注的音频文件对应字段留空。基准真值标注源自原始数据源，具体解析方式详见交付文档D4.4与D4.10的正文内容。 剩余的《SINGLE EVENT - Ground Truth.csv》仅包含以下2列： 1. **Freesound标识符（Freesound ID）**：Freesound平台用于标识音频片段的声音ID。 2. **单事件判定（Single Event:）**：布尔值，用于标识对应音频是否属于单事件音频。单事件标注由交付文档的作者根据交付文档D4.10中的方法采集得到。 ### 音频数据获取方式 本节提供本数据集所用音频文件的获取指南。需注意，受授权协议限制，本仓库无法重新分发大部分基准真值对应的音频文件。 1. **Apple Loops（APPL）**：本数据集包含Apple音乐软件（如Logic Pro、GarageBand）内置的部分音乐循环。获取此类循环需持有对应软件的授权。本数据集的详细配置指南见此处。 2. **Carlos Vaquero乐器数据集（CVAQ）**：本数据集包含Carlos Vaquero在其硕士论文阶段录制的单乐器音频。相关音频以Freesound合集形式发布，可通过以下页面下载：https://freesound.org/people/Carlos_Vaquero/packs 3. **Freesound循环4k数据集（FSL4）**：本数据集包含从Freesound平台选取的部分音乐循环。本数据集的详细配置指南见此处。 4. **Giant Steps调式数据集（GSKY）**：本数据集包含从Beatport平台选取的带调式标注的音频预览片段。音频与原始标注可在此处获取。 5. **Good-sounds数据集（GSND）**：本数据集包含乐器样本的单声道录音。完整说明、原始标注与音频文件均可在此处获取。 6. **爱荷华大学乐器样本数据集（IOWA）**：本数据集由爱荷华大学电子音乐工作室创建，包含乐器样本录音。可通过访问以下网站申请获取本数据集。 7. **Mixcraft循环数据集（MIXL）**：本数据集包含Acoustica旗下Mixcraft音乐软件内置的部分音乐循环。获取此类循环需持有对应软件的授权。本数据集的详细配置指南见此处。 8. **NSynth数据集测试与验证集（NSYT与NSYV）**：NSynth是由Google Magenta团队构建的大规模高质量合成音乐音符标注数据集。完整数据集说明、原始标注与音频文件均可在此处获取。 9. **爱乐乐团声音样本数据集（PHIL）**：本数据集包含由爱乐乐团演奏家录制的数千个免费可下载声音样本。音频文件可直接从爱乐乐团官网免费下载。 10. **Freesound单事件数据集（SINGLE EVENT）**：本数据集包含从Freesound平台选取的部分音频片段，涵盖单事件音频信号与多事件音频信号。可通过CSV文件中提供的ID标识符，从Freesound平台下载对应的原始音频片段，具体操作可参考此处所述流程。