提出一种名为 Top K Entropy 的主动学习不确定性聚合策略,通过优先选择最不确定的音频片段进行标注,仅需 8% 标签即可达到全量数据训练性能,显著提升了生物声学监测场景中的标注效率。
摘要:
声音事件检测(SED)应用中收集的大量音频数据需要高效的标注策略以实现监督学习。人工标注昂贵且耗时,使得主动学习(AL)成为减少标注工作量的有效方法。我们提出了 Top K Entropy,一种用于 AL 的新型不确定性聚合策略,它优先选择音频录音中最不确定的片段,而非对所有片段的不确定性取平均值。该方法能够选择整段录音进行标注,提高了稀疏数据场景下的效率。我们将 Top K Entropy 与随机采样和 Mean Entropy 进行了比较,结果表明更少的标签即可达到相同的模型性能,特别是在声音事件稀疏的数据集中。评估在包含狐獴、狗和婴儿哭声事件的公园录音混合音频上进行,代表了真实的生物声学监测场景。使用 Top K Entropy 进行主动学习,仅需 8% 的标签即可获得与在全标注数据集上训练相当的性能。Top K Entropy 优于 Mean Entropy,表明最好让最不确定的片段代表整个音频文件的不确定性。研究结果突显了 AL 在音频和时间序列应用(包括生物声学)中实现可扩展标注的潜力。
英文摘要:
The vast amounts of audio data collected in Sound Event Detection (SED) applications require efficient annotation strategies to enable supervised learning. Manual labeling is expensive and time-consuming, making Active Learning (AL) a promising approach for reducing annotation effort. We introduce Top K Entropy, a novel uncertainty aggregation strategy for AL that prioritizes the most uncertain segments within an audio recording, instead of averaging uncertainty across all segments. This approach enables the selection of entire recordings for annotation, improving efficiency in sparse data scenarios. We compare Top K Entropy to random sampling and Mean Entropy, and show that fewer labels can lead to the same model performance, particularly in datasets with sparse sound events. Evaluations are conducted on audio mixtures of sound recordings from parks with meerkat, dog, and baby crying sound events, representing real-world bioacoustic monitoring scenarios. Using Top K Entropy for active learning, we can achieve comparable performance to training on the fully labeled dataset with only 8% of the labels. Top K Entropy outperforms Mean Entropy, suggesting that it is best to let the most uncertain segments represent the uncertainty of an audio file. The findings highlight the potential of AL for scalable annotation in audio and time-series applications, including bioacoustics.
Few-shot bioacoustic event detection at the DCASE 2022 challenge
Queen Mary University of London(伦敦玛丽女王大学);Nantes Université(南特大学);École Centrale Nantes(南特中央理工学院);CNRS(法国国家科学研究中心);University of Konstanz(康斯坦茨大学);Max Planck Institute of Animal Behavior(马克斯·普朗克动物行为研究所);BIOTOPIA Naturkundemuseum Bayern(巴伐利亚BIOTOPIA自然历史博物馆);AGH University of Science and Technology(AGH科技大学);Cornell University(康奈尔大学);Tilburg University(蒂尔堡大学);Naturalis Biodiversity Centre(Naturalis生物多样性中心);University Ramon Llull(拉蒙·柳利大学);Oxford University(牛津大学);Syracuse University(雪城大学);University of Salford(索尔福德大学);University of Surrey(萨里大学);Sonantic Limited(Sonantic有限公司)
ForestIR: Physics-Informed Forest Sound Simulation for Array-Based Bioacoustic Remote Sensing
Department of Statistical Science, Duke University(统计科学系,杜克大学);Department of Biological and Environmental Science, University of Jyväskylä(生物与环境科学系,于韦斯屈耶大学);Faculty of Biological and Environmental Sciences, University of Helsinki(生物与环境科学学院,赫尔辛基大学);Organismal and Evolutionary Biology Research Programme, Faculty of Biological and Environmental Sciences, University of Helsinki(生物与进化生物学研究计划,生物与环境科学学院,赫尔辛基大学);Digital Services, University of Jyväskylä(数字服务,于韦斯屈耶大学);Open Science Centre, University of Jyväskylä(开放科学中心,于韦斯屈耶大学)
Full-Stack Bioacoustics: Field Kit to AI to Action (Workshop report)
Naturalis Biodiversity Center(自然生物多样性中心);Tilburg University(蒂尔堡大学);Universiteit van Amsterdam(阿姆斯特丹大学);CODE University of Applied Sciences Berlin(柏林CODE应用科学大学);University of Cambridge(剑桥大学)
Robust sound event detection in bioacoustic sensor networks
Cornell Lab of Ornithology, Cornell University(康奈尔大学康奈尔鸟类学实验室);Music and Audio Research Laboratory, New York University(纽约大学音乐与音频研究实验室);Center for Urban Science and Progress, New York University(纽约大学城市科学与进步中心)
Leveraging tropical reef, bird and unrelated sounds for superior transfer learning in marine bioacoustics
University College London(伦敦大学学院);Google DeepMind(谷歌DeepMind);Zoological Society of London(伦敦动物学会);Conservation Metrics Inc.(保护指标公司);University of New Hampshire(新罕布什尔大学);K. Lisa Yang Center for Conservation Bioacoustics, Cornell Lab of Ornithology, Cornell University(康奈尔大学鸟类学实验室K. Lisa Yang保护生物声学中心);Department of Public and Ecosystem Health, Cornell University(康奈尔大学公共与生态系统健康系);Sound Ocean Science(海洋声学科学);University of Exeter(埃克塞特大学);University of Bristol(布里斯托大学);IPB University(IPB大学);Google Research(谷歌研究院)
Few-shot bioacoustic event detection at the DCASE 2023 challenge
Queen Mary University of London(伦敦玛丽女王大学);Naturalis Biodiversity Centre(Naturalis生物多样性中心);Ramon Llull University(拉蒙·柳利大学);University of Konstanz(康斯坦茨大学);Max Planck Institute of Animal Behavior(马克斯·普朗克动物行为研究所);Oxford University(牛津大学);Syracuse University(雪城大学);University of Surrey(萨里大学);University of Salford(索尔福德大学);Naturkundemuseum Bayern(巴伐利亚自然历史博物馆);AGH University of Science and Technology(AGH科技大学);Nantes Université(南特大学);École Centrale Nantes(南特中央理工学院);CNRS(法国国家科学研究中心);Tilburg University(蒂尔堡大学)
RAVEN X High Performance Data Mining Toolbox for Bioacoustic Data Analysis
Bioacoustics Research Program, Cornell Lab of Ornithology, Cornell University(康奈尔大学鸟类学实验室生物声学研究项目);Department of Computer Science, San Diego State University(圣地亚哥州立大学计算机科学系);US Navy, SPAWAR Systems Center Pacific(美国海军太平洋空间与海战系统中心)
Uncertainty Calibration of Multi-Label Bird Sound Classifiers
INS, Kiel University, Germany(基尔大学);Tilburg University, Netherlands(蒂尔堡大学);Naturalis Biodiversity Center, Netherlands(自然史生物多样性中心);IES, University of Kassel, Germany(卡塞尔大学)
Automatic Identification of Animal Breeds and Species Using Bioacoustics and Artificial Neural Networks
Research Collaboratory for Advanced Intelligent Systems(高级智能系统研究合作中心);Institute of Computer Science, University of the Philippines Los Baños(菲律宾大学洛斯巴诺斯分校计算机科学学院)
Generalization in birdsong classification: impact of transfer learning methods and dataset characteristics
Naturalis Biodiversity Center(自然生物多样性中心);Xeno-Canto Foundation(异音基金会);Landesbund fur Vogel- und Naturschutz(鸟类与自然保护州联合会);Department of Cognitive Science and Artificial Intelligence, Tilburg University(蒂尔堡大学认知科学与人工智能系)
Learning to detect an animal sound from five examples
Queen Mary University of London(伦敦玛丽女王大学);Nantes Universite, Ecole Centrale Nantes, CNRS, LS2N, UMR 6004(南特大学,南特中央理工大学,法国国家科学研究中心,LS2N,UMR 6004);University of Konstanz(康斯坦茨大学);Max Planck Institute of Animal Behavior(马克斯·普朗克动物行为研究所);Universitat Ramon Llull(拉蒙·尤以大学);BIOTOPIA Naturkundemuseum Bayern(巴伐利亚州自然博物馆BIOTOPIA);AGH University of Science and Technology(AGH科技大学);University of Salford(索尔福德大学);University of Surrey(萨里大学);Aarhus University(奥胡斯大学);Syracuse University(雪城大学);Woods Hole Oceanographic Institution(伍兹霍尔海洋研究所)
2-Dimensional spectral gating for denoising bioacoustics recordings
Mila - Quebec AI Institute(米拉-魁北克人工智能研究所);Pioneer Centre for AI, University of Copenhagen(哥本哈根大学人工智能先锋中心);McGill University(麦吉尔大学);Stockholm University(斯德哥尔摩大学)
Whombat: An open-source annotation tool for machine learning development in bioacoustics
University College London(伦敦大学学院);University of Edinburgh(爱丁堡大学);Oregon State University(俄勒冈州立大学);USDA Forest Service(美国农业部林务局);Bat Conservation Trust(蝙蝠保护信托基金)
Proposal-based Few-shot Sound Event Detection for Speech and Environmental Sounds with Perceivers
Allen Institute for Artificial Intelligence(艾伦人工智能研究所);Western Washington University(西华盛顿大学);DocuSign(DocuSign公司);Pacific Northwest National Laboratory(太平洋西北国家实验室)
Western Mediterranean wetlands bird species classification: evaluating small-footprint deep learning approaches on a new annotated dataset
La Salle-Universitat Ramon Llull(拉蒙·柳利大学拉萨尔学院)
部署一个运行在无线声学传感器网络上的专家系统,该网络由生物声学监测设备组成,可以从声音中识别鸟类,这将使许多具有生态价值的任务实现自动化,包括分析鸟类种群组成或检测环境利益领域的濒危物种。得益于人工智能的最新进展,赋予这些设备准确的音频分类功能是可能的,其中深度学习技术表现出色。然而,使生物声学设备负担得起的一个关键问题是使用可以嵌入在资源和电池受限的硬件平台中的小足迹深度神经网络。出于这个原因,这项工作提出了两个重型和大型足迹深度神经网络(VGG 16和ResNet 50)和轻量级替代方案MobileNetV2之间的关键比较分析。我们的实验结果表明,MobileNetV2的平均F1分数比ResNet50低不到5%(0.789 vs. 0.834),表现优于VGG16,占用空间小近40倍。此外,为了比较模型,我们创建并公开了西地中海湿地鸟类数据集,包括201.6分钟和5,795个Aiguamolls de l'Empordà自然公园20种特有鸟类的音频摘录。
Exploring bat song syllable representations in self-supervised audio encoders
Institute for Logic, Language and Computation, University of Amsterdam(阿姆斯特丹大学逻辑、语言与计算研究所);Museum für Naturkunde, Leibniz Institute for Evolution and Biodiversity Science(莱布尼茨进化与生物多样性科学自然博物馆)