[DCASE2022 Task 3] Synthetic SELD mixtures for baseline training

<strong>DESCRIPTION:</strong><br> <br> This audio dataset serves serves as supplementary material for the DCASE2022 Challenge Task 3: Sound Event Localization and Detection Evaluated in Real Spatial Sound Scenes. The dataset consists of synthetic spatial audio mixtures of sound events spatialized for two different spatial formats using real measured room impulse responses (RIRs) measured in various spaces of Tampere University (TAU). The mixtures are generated using the same process as the one used to generate the recordings of the TAU-NIGENS Spatial Sound Scenes 2021 dataset for the DCASE2021 Challenge Task 3. The SELD task setup in DCASE2022 is based on spatial recordings of real scenes, captured in the STARS22 dataset. Since the task setup allows use of external data, these synthetic mixtures serve as additional training material for the baseline model, and they are shared for reasons of reproducibility. For more details on the task setup, please refer to the task description. Note that the generator code and the collection of room responses used to spatialize sound samples will be also be made available soon. For more details on the recording of RIRs, spatialization, and generation, see: Archontis Politis, Sharath Adavanne, Daniel Krause, Antoine Deleforge, Prerak Srivastava, Tuomas Virtanen (2021). A Dataset of Dynamic Reverberant Sound Scenes with Directional Interferers for Sound Event Localization and Detection. In <em>Proceedings of the Detection and Classification of Acoustic Scenes and Events 2020 Workshop (DCASE2021)</em>, Barcelona, Spain. available here. <strong>SPECIFICATIONS:</strong> <strong>13 target sound classes</strong> (see task description for details) The sound event samples are sources from the <strong>FSD50K</strong> dataset, based on affinity of the labels in that dataset to the target classes. The selection on distinguishing which labels in FSD50K corresponded to the target ones, then selecting samples that were tagged with only those labels, and additionally that they had annotator rating of Present and Predominant (see FSD50K for more details). The list of the selected files is included here. <strong>1200</strong> 1-minute long spatial recordings Sampling rate of<strong> 24kHz</strong> Two 4-channel recording formats, first-order Ambisonics (<strong>FOA</strong>) and tetrahedral microphone array (<strong>MIC</strong>) Spatial events spatialized in <strong>9 unique rooms</strong>, using measured RIRs for the two formats Maximum <strong>polyphony of 2</strong> (with possible same-class events overlapping) Even though the whole set is used for training of the baseline without distinction between the mixtures, we have included a <strong>separation into a training and testing split</strong>, in case on one needs to test the performance purely on those synthetic conditions (for example for comparisons with training on mixed synthetic-real data, fine-tuning on real data, or training on real data only). The training split is indicated as <strong>fold1</strong> in the dataset, contains 900 recordings spatialized on 6 rooms (150 recordings/room) and it is based on samples from the development set of FSD50K. The testing split is indicated as <strong>fold2</strong> in the dataset, contains 300 recordings spatialized on 3 rooms (100 recordings/room) and it is based on samples from the evaluation set of FSD50K. Common metadata files for both formats are provided. For the file naming and the metadata format, refer to the task setup. <strong>FSD50K SELECTION:</strong> The list of selected sound event recordings is included along the recordings and metadata, as <strong>FSD50K_selected.txt</strong>. Each line in the text file has the following structure: <pre><code>[target_label]/[train/test]/[FSD50K_label]/filename.wav</code></pre> with an example: <pre><code>domesticSounds/train/Boiling/16584.wav</code></pre> meaning that the file 16584.wav from FSD50K, with the <em>Boiling</em> label of FSD50K, is included in the samples for the training split of those synthetic recordings, and it is mapped to the target class of <em>domestic sounds. </em>Note that there can be multiple FSD50K labels mapped the same target class. Also note that if these are downloaded from FSD50K, and a folder structure is created that replicates the structure in the list, the resulting folder can be used out-of-the-box with the scene generator to generate new mixtures with the same or different parameters. Note that no sounds form FSD50K have been selected for the <em>Music</em> target class. Background and pop music tracks from the public domain have been cropped and used instead. <strong>DOWNLOAD INSTRUCTIONS:</strong> Download the zip files and use your preferred compression tool to unzip these split zip files. To extract a split zip archive (named as zip, z01, z02, ...), you could use, for example, the following syntax in Linux or OSX terminal: Combine the split archive to a single archive: <pre>zip -s 0 split.zip --out single.zip</pre> Extract the single archive using unzip: <pre>unzip single.zip</pre><strong>数据集说明：</strong><br> <br> 本音频数据集作为DCASE2022挑战赛任务3：真实空间声音场景下的声音事件定位与检测的补充材料。该数据集包含两类空间格式的合成空间声音事件混合信号，其空间化处理采用坦佩雷大学（Tampere University, TAU）多个场景中实测得到的房间冲激响应（room impulse responses, RIRs）。本数据集的生成流程与DCASE2021挑战赛任务3所用的TAU-NIGENS 2021空间声音场景数据集的录制生成流程一致。DCASE2022的声音事件定位与检测（SELD）任务设置基于STARS22数据集中的真实场景空间录音。由于该任务允许使用外部数据，此类合成混合信号可作为基线模型的额外训练素材，且为保障实验可复现性而公开共享。如需了解任务设置的更多细节，请参阅任务说明文档。需注意，用于声音样本空间化处理的生成代码与房间响应集也将尽快公开。若需了解房间冲激响应录制、空间化处理及数据生成的更多细节，请参阅：Archontis Politis, Sharath Adavanne, Daniel Krause, Antoine Deleforge, Prerak Srivastava, Tuomas Virtanen (2021). 面向声音事件定位与检测的带定向干扰源的动态混响声音场景数据集。收录于<em>2020年声音场景检测与分类研讨会论文集（DCASE2021）</em>，西班牙巴塞罗那。可在此处获取。<br><br><strong>数据集规格：</strong><br> <strong>13个目标声音类别</strong>（详细类别清单请参阅任务说明）。<br>声音事件样本取自<strong>FSD50K</strong>数据集，选取依据为该数据集标签与目标类别的匹配度。筛选流程为：先确定FSD50K中与目标类别对应的标签，再仅选取带有这些标签的样本，且要求标注者评分标记为Present（存在）与Predominant（主导）（详情请参阅FSD50K文档）。所选文件清单可在此处获取。<br><strong>1200条</strong>时长为1分钟的空间录音，采样率为<strong>24kHz</strong>。<br>包含两种4通道录音格式：一阶Ambisonics（first-order Ambisonics, FOA）与四面体麦克风阵列（tetrahedral microphone array, MIC）。<br>空间事件在<strong>9个独立房间</strong>中完成空间化处理，采用针对两种格式实测得到的房间冲激响应。<br>最大<strong>复音数为2</strong>（允许同类事件重叠）。<br>尽管完整数据集无需区分混合信号即可用于基线模型训练，但我们仍将其划分为<strong>训练集与测试集划分</strong>，以供研究者仅在合成条件下测试模型性能（例如，用于对比基于合成-真实混合数据训练、基于真实数据微调或仅基于真实数据训练的模型效果）。训练集在数据集中标记为<strong>fold1</strong>，包含分布于6个房间的900条录音（每个房间150条），样本取自FSD50K的开发集。测试集标记为<strong>fold2</strong>，包含分布于3个房间的300条录音（每个房间100条），样本取自FSD50K的评估集。两种格式共用元数据文件。如需了解文件命名与元数据格式，请参阅任务设置文档。<br><br><strong>FSD50K样本筛选：</strong><br>所选声音事件录音清单与录音文件、元数据一同提供，文件名为<strong>FSD50K_selected.txt</strong>。文本文件中每一行的格式如下：<br><pre><code>[target_label]/[train/test]/[FSD50K_label]/filename.wav</code></pre><br>示例如下：<br><pre><code>domesticSounds/train/Boiling/16584.wav</code></pre><br>该示例表示，来自FSD50K的16584.wav文件，其FSD50K标签为Boiling，被纳入该合成录音的训练集样本，且映射至domestic sounds（室内声音）这一目标类别。需注意，多个FSD50K标签可映射至同一目标类别。此外，若从FSD50K下载对应文件并按照清单中的目录结构创建文件夹，可直接使用场景生成工具生成参数相同或不同的新混合信号。需注意，<em>Music（音乐）</em>目标类别未选取任何FSD50K的声音样本，取而代之的是裁剪自公共领域的背景音与流行音乐曲目。<br><br><strong>下载说明：</strong><br>请下载分卷压缩包并使用任意解压工具解压。若需在Linux或OSX终端中合并分卷压缩包，可使用以下命令：<br>将分卷归档合并为单归档文件：<br><pre>zip -s 0 split.zip --out single.zip</pre><br>使用unzip解压单归档文件：<br><pre>unzip single.zip</pre>