simAugCENSE-18K

Created By Félix Gontier and Mathieu Lagrange, LS2N, CNRS, Ecole Centrale Nantes Contact : [e-mail] If used for research, please refer to: <pre>@article{gontier2021spatial, title={Spatial and temporal pretext tasks for few-shot urban sound classification}, author={Félix Gontier and Vincent Lostanlen, and Mathieu Lagrange and Nicolas Fortin and Jean-Francois Petiot and Catherine Lavandier}, journal={submitted}, year={2021} } </pre> simAugCENSE-18k contains 400 acoustic scenes of duration equal to 45 seconds.<br> We synthesized these polyphonic scenes via the simScene software, based on monophonic audio clips from the CENSE-2k dataset.<br> The total duration of the dataset is equal to 18k seconds, i.e., five hours.The audio is made available as third octave spectral data, see demoTob.zip for an implementation of its computation from audio in Python. &gt;&gt; import numpy as np &gt;&gt; s=np.load('simAugCENSE-18k_train_spectralData.npy') &gt;&gt; print(s.shape) (280, 351, 29) The three dimensions respectively corresponds to the sceneId, the frameId (time), and the spectralId (frequency). &gt;&gt; a=np.load('simAugCENSE-18k_train_annotations.npy') &gt;&gt; print(a.shape) (280, 344, 3) &gt;&gt; s=np.load('simAugCENSE-18k_validation_spectralData.npy') &gt;&gt; print(s.shape) (120, 344, 29) &gt;&gt; a=np.load('simAugCENSE-18k_validation_annotations.npy') &gt;&gt; print(a.shape) (120, 344, 3) The 3 dimensions corresponds to the sceneId, the frameId (time), the sourceId (traffic, voice, birds). Annotation is provided as a binary indicator of source presence for one second, that is 8 consecutive 125 ms frames with a hop of one frame.本数据集由Félix Gontier与Mathieu Lagrange创建，依托法国国家科学研究中心（CNRS）南特中央理工学院LS2N实验室完成。联系方式：[e-mail]。若用于学术研究，请引用如下文献： bibtex @article{gontier2021spatial, title={Spatial and temporal pretext tasks for few-shot urban sound classification}, author={Félix Gontier and Vincent Lostanlen, and Mathieu Lagrange and Nicolas Fortin and Jean-Francois Petiot and Catherine Lavandier}, journal={submitted}, year={2021} } 其中论文标题可译为《面向少样本（few-shot）城市声分类的时空预训练任务》。 simAugCENSE-18k数据集包含400段时长为45秒的声学场景。我们基于CENSE-2k数据集的单声道音频片段，通过simScene软件合成了这些多声部声学场景。本数据集总时长为18000秒，即五小时。 本数据集的音频以三分之一倍频程频谱数据形式提供，相关Python计算实现可参考demoTob.zip压缩包。示例代码如下： >>> import numpy as np >>> s=np.load('simAugCENSE-18k_train_spectralData.npy') >>> print(s.shape) (280, 351, 29) 该数组的三个维度依次对应场景ID（sceneId）、帧ID（frameId，代表时间维度）与频谱ID（spectralId，代表频率维度）。 标注数据相关示例代码如下： >>> a=np.load('simAugCENSE-18k_train_annotations.npy') >>> print(a.shape) (280, 344, 3) >>> s=np.load('simAugCENSE-18k_validation_spectralData.npy') >>> print(s.shape) (120, 344, 29) >>> a=np.load('simAugCENSE-18k_validation_annotations.npy') >>> print(a.shape) (120, 344, 3) 标注数组的三个维度依次对应场景ID、帧ID（frameId，时间维度）与声源ID（sourceId，涵盖交通、人声、鸟鸣三类声源）。标注以二进制指示符形式给出，用于表征对应1秒时长内是否存在指定声源：每1秒时长包含8个连续的125毫秒帧，且帧移为1帧。