HearTheSpecies: Using computer audition to understand the drivers of soundscape composition, and to predict parasitation rates based on vocalisations of bird species

heartheespecies是奥格斯堡大学医疗保健和福利嵌入式智能主席（EIHW）和弗莱堡大学地球植物学主席（ALU-FR）的跨学科项目。土地利用的集约化是当前生物多样性丧失的主要原因之一。为了更好地了解和监测土地利用强度与生物多样性丧失之间的联系，heartheespecies旨在利用迄今尚未得到充分研究的数据池的潜力：音频。土地利用对景观声景观的影响表现在不同的方面和尺度上。发声动物群落的物种消失或景观结构和植被密度的变化会改变生物声（野生动物的声音）、地质声（非生物自然界的声音）和人声（人类引起的声音）的组成。在更精细的尺度上，土地利用甚至可以通过影响适应性和寄生率来影响单个动物的声音特征。在生物多样性探索区内建立的土地利用梯度为研究这些关系提供了一个很好的研究平台，从而建立基于人工智能的自主声学监测工作流程。具体来说，作为heartheespecies的一部分，我们的目标是开发基于人工智能的自动记录和分离工具，可以从交织的音景记录中粗分离生物声音、人声和地声，以及细粒度检测和分离物种和特定的非生物声音。为此，我们将注释以前在生物多样性探索项目中收集音频记录的现有数据，以实现人工智能算法的训练，并在联合多站点实验REX和FOX中收集新的音频数据。下一步，我们将利用这些分离的声音来模拟当地和区域土地利用强度、景观配置和植被结构对声景观组成和声学群落个体物种的影响，并根据它们的鸣叫特征预测鸟类的寄生率。

HearTheSpecies is an interdisciplinary project of the Chair of Embedded Intelligence for Health Care and Wellbeing (EIHW) at the University of Augsburg and the Chair of Geobotany at the University of Freiburg (ALU-FR). The intensification of land use is one of the main causes of the current loss of biodiversity. In order to better understand and monitor the links between land use intensity and biodiversity loss, HearTheSpecies aims to harness the potential of a hitherto under-researched data pool: Audio. The impact of land use on the soundscape of a landscape manifests itself in different aspects and scales. Species loss of vocalising animal communities or changes in landscape structure and vegetation density alter the composition of biophony (sounds of wildlife), geophony (sounds of abiotic nature) and anthropophony (sounds caused by humans). On a much finer scale, land use can even affect the vocal characteristics of individual animals by influencing fitness and parasitisation rates. The established land use gradient within biodiversity exploratories provides an excellent research platform to investigate these relationships and thus establish AI-based autonomous acoustic monitoring workflows. Specifically, as part of HearTheSpecies, we aim to develop AI-based automatic diarisation and separation tools that enable coarse separation of biophony, anthropophony and geophony from interwoven soundscape recordings, and fine-grained detection and separation of species and specific abiotic sounds. To do this, we will annotate existing data from previous projects that collected audio recordings within the Biodiversity Exploratories to enable the training of AI algorithms, and collect new audio data in the joint multi-site experiments REX and FOX. In a next step, we will use these separated sounds to model the effects of local and regional land use intensity, landscape configuration and vegetation structure on the composition of the soundscape and individual species of the acoustic community, and predict parasitisation rates in birds based on their song characteristics.