3D audio techniques for acoustic monitoring of rainforest biodiversity

声音携带着当地生物多样性的大量信息，被广泛的分类群用于导航和交流。声学信息现在通常用于协助对许多这些物种的点调查，帮助识别蝙蝠，蚱蜢，鸟类，两栖动物，甚至单个动物。集成的3D音频传感器可以提供额外的线索，不仅可以识别物种和数量，还可以收集有关运动、方向、速度和其他相关数据的信息。3D麦克风阵列将被设计并集成在定制的多声道录音机中。高质量的低功率驻极体麦克风将被安排在一个球形阵列中，以便能够录制全向音频线索，以及定向音频线索。整个系统将集成太阳能深循环电池，用于定期上传记录数据的GPRS/3G/4G连接，以及数据压缩能力。该系统将被设计成允许简单的一人安装在采油树上。将进行一系列基准测试，以校准记录质量和数据压缩，允许记录有意义和可用的数据，同时允许或定期上传非最佳网络覆盖的记录数据。该系统将首先在英国进行测试，使用戴森设计工程学院的VR声学设施和Silwood Park的现场站点，然后将部署在马来西亚，作为Ewers教授研究小组持续监测的一部分。有了记录的数据，以前开发的统计方法将用于实施自主的、连续的生物多样性监测。首先，声学信号中的物种呼叫将使用现有的信号处理技术自动检测，该技术允许将超过9000种信号处理算法应用于物种呼叫，以开发物种的声学“指纹”（将寻求帝国数学学院的尼克琼斯博士的支持，用于这个特定的研究阶段）。其次，空间音频数据将使用波束成形和3D Ambisonic等技术进行解码，以确定检测到的声音和呼叫的角度位置和距离。第三，我们将使用这些数据来识别声音记录中的个体，并使用这些数据应用可探测性统计来获得更准确的物种丰度测量。这些指标将使用蝙蝠、鸟类、两栖动物、哺乳动物和无脊椎动物物种丰富度的实地数据进行校准，这些数据是尤厄斯教授在马来西亚的研究小组在持续监测中收集的。每个声学监测点的各种分类群多样性快照样本将在多个时间点使用，以测试观察到的生物多样性与声学衍生物种记录和信息度量之间的相关性。这些校准的指标将在原始雨林和被砍伐雨林中进行比较，并沿着历史伐木强度的梯度进行比较，以确定森林干扰对生物多样性的影响。拟议的研究建立在当前的博士项目基础上（由同一团队监督），旨在整合已开发的声学监测设备3D麦克风阵列和编码器，并实施新的方法和技术来解码和分析空间声学数据。

Sound carries substantial information about local biodiversity, being used for navigation and communication by a wide range of taxa. Acoustic information is now commonly used to assist in point-surveys of many of these species, aiding the identification of bats, grasshoppers, birds, amphibians, and even individual animals. The integration of 3D audio sensors for acoustic monitoring could offer additional cues in order not only to be able to identify the species and numbers, but also to gather information about movements, direction, velocity, and other relevant data. A 3D microphone array will be designed and integrated within a custom built multichannel audio recorder. High-quality low-power electret microphones will be arranged in a spherical array, in order to allow the recording of omnidirectional audio cues, as well as directional ones. The whole system will be integrated with a solar-powered deep-cycle battery, GPRS/3G/4G connectivity for regularly uploading recorded data, and data compression capabilities.The system will designed in order to allow for simple one-manned installation on a tree. A series of benchmark tests will be carried out in order to calibrate the recording quality and data compression, allowing the recording of meaningful and usable data, and at the same time giving the possibility or regularly upload the recording data with non-optimal network coverage. The system will be initially tested in the UK, using the VR acoustic facilities within the Dyson School of Design Engineering and the field sites at Silwood Park, and will then be deployed in Malaysia, as part of the on-going monitoring by Prof Ewers' research group.With the recorded data, previously developed statistical methods will be used to implement autonomous, continuous biodiversity monitoring. First, species calls in the acoustic signal will be automatically detected using existing signal processing techniques, which allow to apply a battery of more than 9000 signal processing algorithms to species' calls to develop an acoustic 'fingerprint' for species (support from Dr Nick Jones from Imperial Mathematics will be sought for this specific research stage). Second, the spatial audio data will be decoded using techniques such as beamforming and 3D Ambisonic, to determine angular position and distance of detected sounds and calls. Third, we will use these data to identify individuals within an acoustic record, and use those data to apply detectability statistics to gain more accurate measures of species' abundances.These metrics will be calibrated using field data on the species richness of bats, birds, amphibians, mammals and invertebrates, collected as part of on-going monitoring by Prof Ewers' research group in Malaysia. Snapshot samples of diversity of the various taxa at each of the acoustic monitoring sites will be used at multiple time points to test the correlations between observed biodiversity and acoustically derived species records and information metrics. These calibrated metrics will be compared in primary and logged rainforest and along a gradient of historical logging intensity to determine the impacts of forest disturbance on biodiversity.The proposed research builds on a current PhD project (supervised by the same team), and aims at integrating on the developed acoustic monitoring device 3D microphone arrays and encoders, and at implementing novel methods and techniques to decode and analyse the spatial acoustic data.