AniOaks/drone-audio-detection-samples

Drone Audio Detection Samples (DADS) 是目前最大的公开无人机音频数据库，专门用于基于深度学习技术开发无人机检测系统。所有音频文件标准化为16,000 Hz采样率、16位深度、单声道，长度从500毫秒到几分钟不等。大多数无人机音频文件经过手动修剪，以确保录音中始终存在无人机声音；但部分数据集包含从远距离麦克风捕获的无人机音频，需要多次仔细聆听才能检测背景中的无人机声音。数据集结构分为两类：0（无无人机）包含无无人机声音的音频样本，1（有无人机）包含各种无人机的音频样本。统计显示，无无人机类别有16,729个文件，总时长121,714秒（约33.81小时），平均时长7.28秒；有无人机类别有163,591个文件，总时长97,563.39秒（约27.10小时），平均时长0.60秒。所有音频文件共享以下属性：采样率16 kHz、位深度16位、单声道、WAV PCM 16位格式。该数据集旨在用于安全、监控及相关研究领域开发鲁棒的基于音频的无人机检测算法和模型。数据来源包括多个无人机音频数据集（如Drone Audio Dataset、SPCup 19 Egonoise Dataset等）和非无人机音频数据集（如Urban Sound 8K、TUT Acoustic Scenes 2017等）。许可为MIT许可证。Drone Audio Detection Samples (DADS) is currently the largest publicly available drone audio database, specifically designed for developing drone detection systems based on deep learning technologies. All audio files are standardized to a sampling rate of 16,000 Hz, 16-bit depth, and mono audio, with durations ranging from 500 milliseconds to several minutes. Most drone audio files have been manually trimmed to ensure that drone sounds are consistently present in the recordings; however, some subsets of the dataset contain drone audio captured by distant microphones, which requires multiple careful listens to detect the background drone sounds. The dataset is structured into two categories: 0 (no drone) for audio samples without any drone sounds, and 1 (with drone) for audio samples of various drones. Statistics show that the no-drone category contains 16,729 files, with a total duration of 121,714 seconds (approximately 33.81 hours) and an average duration of 7.28 seconds; the with-drone category contains 163,591 files, with a total duration of 97,563.39 seconds (approximately 27.10 hours) and an average duration of 0.60 seconds. All audio files share the following attributes: 16 kHz sampling rate, 16-bit bit depth, mono audio, and WAV PCM 16-bit format. This dataset is intended for developing robust audio-based drone detection algorithms and models in the fields of security, surveillance and related research. Its data sources include multiple drone audio datasets (such as Drone Audio Dataset, SPCup 19 Egonoise Dataset, etc.) and non-drone audio datasets (such as Urban Sound 8K, TUT Acoustic Scenes 2017, etc.). The dataset is licensed under the MIT License.