面向鸟鸣声识别任务的深度学习技术

在生态系统中,鸟类是重要的组成部分,对调节生态环境和监测生物多样性至关重要,甚至可以通过监测鸟群动向与监听鸟群异常鸣声对地震、海啸等自然灾害进行辅助预测和防范,为此,鸟鸣声识别和异常鸣声监测成为热门的研究方向。然而,由于传统鸟鸣声识别方法存在特征提取不充分等问题,导致识别率不高。本文采用融合特征的方法结合深度学习技术提取鸟鸣声特征,融合特征选择改良后的对数梅尔谱差分参数同原始信号参数拼接所得的特征;深度学习方法是基于Dense Net121网络结构,并融入自注意力模块与中心损失函数进行鸟鸣声识别。自注意力模块部分提高了关键通道的特征表达能力;中心损失函数可解决类内特征不紧凑问题。我们通过消融实验对比验证,对在Xeno-Canto世界野生鸟类声音公开数据集上选取的10种鸟类声音进行识别,准确率达到96.9%。代码已开源至Github:https://github.com/Carrie X6/-Xeno-Canto-.git。     

基于声音指数的鸟类多样性监测

鸟类运动能力强,且对生境变化敏感,常作为环境指示物种。鸣声是鸟类重要的通讯方式,其特征具有物种特异性。随着野外录音及鸣声分析技术和设备的快速发展,从录音中自动提取反映鸣声特征的参数,快速评估鸟种多样性已成为可能。相比于传统的鸟类调查方法,通过录音监测鸟类,具有方便快捷的优点。对鸣声反映鸟类多样性的理论基础、常用的声音指数,以及该领域的关键技术加以介绍,为鸟类监测提供方法借鉴。     

中国43种蛙类鸣声特征数据集

无尾两栖动物的鸣声通常具有物种特异性,了解其鸣声特征信息,是利用生物声学进行物种多样性调查及物种监测的前提。本文汇总、整理了2012–2020年间利用高保真录音设备在野外记录的43种(隶属于7科26属)无尾两栖动物的鸣声数据,以及相应的鸣声采集信息。对音频文件进行降噪处理后,提供了由61个鸣声的波形图及语图组成的鸣声特征数据集。本数据集展示了鸣声的多种时域和频域信息,如单音节或多音节、音节数、音节时长、音节间隔、鸣声时长、主频、基频、谐波等,为我国无尾两栖类的声学研究、物种多样性调查及鸣声监测提供了数据支持。     

天行长臂猿鸣唱的声谱特征与性别差异

白眉长臂猿属(Hoolock)物种声音长期被认为不具有性二型,但是识别其声音的性别差异有助于该属物种的种群动态监测与行为学研究。本研究以天行长臂猿(Hoolock tianxing)作为研究对象,标注了11个个体的鸣声数据。我们通过k-中心点聚类(k-medoids聚类)识别音节类型,基于音节组成和音句长度划分音句类型,并且识别了音节特征和音句使用上的性别差异。本研究识别出wa、oo、whoop、ow和eek 5种音节类型以及7种音句类型。雌性的wa和ow音节频率变化的幅度与速率比雄性更大;oo音节的各项频率指标比雄性更低;whoop音节频率变化的幅度与速率相比于雄性更小。在音句的使用上,独猿个体鸣叫的音句与合唱时各自性别贡献的部分更为相似,雄性极少唱主要由whoop音节与ow音节组成的音句G,雌性极少唱音句B (wa-whoop)和C (wa-oo-wawhoop)。本研究表明天行长臂猿的声音中存在性二型,该性别差异不仅有助于对天行长臂猿的监测,而且有助于理解不同声音类型及性别差异的功能。     

基于多传感器的声景大数据在线采集与分析系统研究

声景包含重要的生态信息,具有实时性强、信息密度高的特点,有重要研究价值。现有的声景研究中,音频及相关环境参数采集和分析仍需要大量的人工作业,耗时耗力。基于多传感集成、边缘计算和深度学习技术,建立了一套声景大数据在线采集与分析系统,包括边缘计算节点和中心计算服务器。并通过3个实验站点,进行了近1年的技术验证,实现了声景大数据的自动化在线采集、传输和分析。该系统能适应户外恶劣的自然环境,能根据任务需求持续不断地进行声景大数据在线采集和分析,稳定性好。声学指数可以反映声景变化,但因指数侧重点不同,不同的声学指数之间变化特征差异较大,需要组合使用。通过声纹特征图能直观地识别出不同发声源,对物种的快速识别、声源的分类等具有较强的借鉴意义。系统借助VGGish网络提取的高维声景特征图能很好地识别不同站点和不同时间的声景变化,在不同站点和昼夜上具有较高的区分精度,有快速和直观地反映不同生态系统的类型特征、生态系统动态变化的潜力。丰富声纹特征库、优化声景特征分析神经网络、建设声景长期监测共享网络,有助于扩展系统在物种识别、生物多样性快速分析、生物与环境相互作用机制方面的应用。研究为声景大数据的在线采集...     

声学指数在城市森林鸟类多样性评估中的应用

鸣声是鸟类之间进行沟通和传递信息的重要方式,这为通过声学监测评估鸟类多样性提供了独特的机会。利用声学指数快速评估生物多样性是一种新兴的调查方法,但城市森林中的复杂声环境可能会导致声学指数的指示结果出现偏差。为了解声学指数在城市森林中应用的可行性,本研究在北京市东郊森林公园设置了50个矩阵式调查样点,于2021年4–6月每月进行1次鸟类传统观测和同步鸣声采集,通过比较两种方法的结果来探究声学监测的有效性。采用Spearman相关分析和广义线性混合模型评估6个常用声学指数与鸟类丰富度和多度的关系,并衡量了每个指数的性能。结果表明:(1)本研究共记录到鸟类10目23科35种,通过声学监听识别的总物种数与传统鸟类观测相等,但具体鸟种存在差异;(2)不同月份间声学指数与鸟类丰富度和多度的相关性有明显差别,声学复杂度指数(ACI)和标准化声景差异指数(NDSI)优于其他指数,是评估鸟类多样性的关键变量;(3)声学指数对鸟类多度的预测能力(R²m=0.32,R2c=0.80)要高于丰富度(R²m=0.12,R2c=0.18)。声学指数为快速评估生物多样性提...     

北京常见鸟类鸣声特征分析及其在驱鸟技术中的应用

针对民航业、农业中的人鸟冲突问题,在保护鸟类的前提下,以鸟声驱赶设备为代表的无伤害驱鸟技术成为近年的研究热点.收集北京常见的鸟类--喜鹊Pica pica、灰喜鹊Cyanopica cyana的正常鸣声、受惊鸣声及其天敌苍鹰Accipiter gentilis的正常鸣声进行研究.利用声谱分析方法测量音节的持续时间(Ds)、主能峰频率(Fhp)、基频(Ff)、组成句子的音节个数(Ns.v)和句型等声学特征参数.发现北京地区所采集上述鸟种的鸣声句法结构简单,句型为相同音节的不断重复.在利用鸣声录音制作鸟声驱鸟设备时,此类鸟声便于编辑、修改和使用,并可通过变换多种方式提高使用效果,从而为缓解人鸟冲突提供更多无伤害方法建议,更好地保护鸟类.     

鸟类鸣唱的复杂性和多样性

鸣唱是一种非常重要的鸟类行为学特征。文章通过对大杜鹃Cuculus canorus、赫氏杜鹃C.horsfieldi、红喉蜂鸟Calypte anna、纵纹腹小XiaoAthene noctua、黑卷尾Dicrurus macrocercus、画眉Garrulax canorus和东方大苇莺Acrocephalus orientalis繁殖期鸣唱的频谱分析,初步总结出了鸟类鸣唱复杂性和多样性的一些特点。初步在鸟类鸣唱型、音段和音节等不同层次上总结出6种不同模式：单音节重复型,双音节重复型,单音节鸣唱型,音段序列稳定变化型,多音节序列不稳定变化型和音节音段混合序列不稳定变化型。鸣唱型模式类型的归纳总结,为鸟类物种行为多样性和鸟声复杂性与多样性研究奠定理论基础。     

基于红外相机网络促进我国鸟类多样性监测:现状、问题与前景

近20年来,红外相机调查技术在我国生物多样性监测与野生动物研究中得到了广泛应用。已有的红外相机调查不仅关注哺乳动物类群,而且也记录到了大量鸟类物种,但大多被作为兽类监测中的兼捕(by-catch)记录。我们系统检索并收集了1992年以来,在我国使用红外相机技术的野生动物监测与研究所发表的学术文献、会议报告、新闻报道和部分未发表数据集共230篇(份),从中提取并汇总鸟类物种与分布记录。结果显示,全国通过红外相机技术共记录到至少393个鸟种,分属17目56科,占全国鸟类物种总数的28.67%,其中雀形目物种数最多(268种)。在科的水平上,记录到物种数最多的分别是鸫科(58种)、画眉科(50种)与雉科(42种);另有23科各仅记录到1个物种。在物种数及探测数方面,地面及林下层活动的森林鸟类均是红外相机记录到的绝对优势类群。已发表的红外相机鸟类记录具有区域性不均衡的特征,四川(16个)和云南(14个)是红外相机调查点最多的省区,而四川(160种)、云南(91种)和浙江(66种)则是记录到鸟类物种数最多的省区。据不完全统计,红外相机共记录到区域性鸟类物种新记录104种(次)。考虑到仍有大量红外相机调查中的鸟类记录被忽视或未及发表报道,我国红外相机所记录到的实际鸟类物种多样性应该更高。这些结果表明,红外相机技术在我国鸟类多样性监测和区域性编目工作中具有重要的作用,可以提供高精度、高质量和大数据量的鸟类物种分布数据。对于以鸡形目为代表的地栖鸟类,可以作为目标类群之一纳入现有的基于红外相机技术的标准化长期监测体系,而这样的监测体系也可以为其他鸟类类群的多样性编目和监测提供数据补充和支持。     

沼水蛙繁殖期鸣声特征及鸣叫节律

鸣叫对无尾两栖类动物的生存与繁殖起重要作用。蛙类的鸣叫行为受到环境因素影响表现出一定的节律性。2016年8和9月,采用录音机和指向性话筒,在野外录制了57只沼水蛙（Hylarana guentheri）的鸣声并对其鸣声特征进行分析;通过悬挂录音笔和自动温湿度记录仪研究了沼水蛙鸣叫节律（17 d）及其与环境温度、相对湿度的关系。结果显示,沼水蛙的鸣声由1 ~ 4个音节组成,不同类型鸣声间的音节主频、音节时长存在显著差异（P < 0.05）。该物种全天具有鸣叫行为,13:00 ~ 14:00时为鸣叫高峰期。白天单音节鸣声、双音节鸣声、三音节鸣声、总鸣声和总音节的数量较夜晚显著增加（P < 0.01）。鸣声数量和音节数量均与环境温度呈正相关（P < 0.01）。结果表明,沼水蛙通过改变音节数量、音节主频和音节时长改变鸣叫策略。沼水蛙的鸣叫行为具有昼夜节律性且受环境温度的影响。     

Body size correlates negatively with the frequency of distress calls and songs of Neotropical birds

ABSTRACT The allometric relationship between body size and song frequency has been established in previous studies of temperate and tropical bird communities. However, the relationship between body size and the frequency of distress calls has been examined in only one study of temperate birds. We examined size‐frequency relationships in the distress calls and songs of a Neotropical bird community in northwestern Costa Rica. In 2008 and 2009, we recorded distress calls and determined the body mass of 54 mist‐netted birds representing 38 species, 35 genera, and 14 families. We obtained songs for these same species from sound libraries and commercially available compact discs. For each vocalization, we measured minimum frequency and frequency of maximum amplitude. Larger birds produced lower‐frequency distress calls and songs than smaller birds. Phylogenetically controlled analyses revealed that the frequency of maximum amplitude was negatively correlated with body mass for both distress calls and songs. Minimum frequency was negatively correlated with mass for distress calls, but not songs. Our analyses suggest that the influence of phylogeny on the relationship between frequency characteristics and body size is modest. Pair‐wise comparisons across 37 species revealed that distress calls and songs had similar minimum frequencies, but songs had significantly lower frequencies of maximum amplitude than distress calls. This difference may arise from differences in signal function. Lower‐frequency sounds should transmit farther through forest habitats and songs must often transmit longer distances to reach their intended audience than distress calls. Our results support the general theory that body size is negatively correlated with the frequency of acoustic signals by demonstrating that this pattern holds true for both distress calls and songs in a Neotropical bird community.     

Super-normal length song preferences of female zebra finches (Taeniopygia guttata) and a theory of the evolution of bird song

Two way choice tests show a preference of female zebra finches for male songs four standard deviations longer than normal song. Further tests show the ontogeny of this preference to parallel song learning in general as well as a preference for songs with entirely heterogeneous notes compared to songs with four note repeats. These findings are discussed in relation to a theory of the evolution of bird song from bird calls due to female preferences for longer, more complex vocalizations. This revised version was published online in July 2006 with corrections to the Cover Date.     

White storks Ciconia ciconia eavesdrop on mating calls of moor frogs Rana arvalis

Mating calls of animals are often detected by unintended receivers which use sexual signals to obtain information about the signaller. We investigated whether white storks Ciconia ciconia can eavesdrop mating calls of moor frogs Rana arvalis . White storks are dependent on moor frog abundance in early breeding season. Interspecific eavesdropping by predators is common and well documented in tropical anurans, whereas it is less known in temperate zone. We compared the frequency of approaches of white storks to loudspeakers when frog calls and the song thrush Turdus philomelos songs were simultaneously played back using the later as controls. The loudspeaker broadcasting the calls of male moor frogs clearly attracted white storks at 22 out of 84 nests. The bird songs attracted white storks in only one case. In 19 cases birds left the nest for unknown reasons which were considered as potential foraging movements. The results of this field experiment report a new case of eavesdropping on acoustic signals showing that advertisement calls of temperate moor frogs are an important stimulus for white storks.     

Vogelstimmen

Avian vocalizations Avian vocalizations are part of biodiversity. They are involved in species formation. In many cases, bird songs are known to develop by imitation learning from a species‐specific model and are transmitted between generations of singers. By this way, also dialects can be formed. Bird sounds can be transformed into visual patterns called sonograms. Recognizing songs, calls and instrumental sounds of birds in the wild is a difficult task for many people. With the help of sonograms and other media this task can be overcome more easily. Knowing and discovering bird sounds is a pleasure and a cognitive hobby for an increasing number of people.     

Individuality in Corncrake Crex crex vocalizations

Vocal individuality has potential for the monitoring of bird species that are otherwise difficult to observe. In this study we assessed the potential of this technique for the monitoring of the globally threatened Corncrake Crex crex based on the analysis of calls recorded in Ireland and Scotland in 1993 and 1994. Discriminant function analysis correctly attributed >80% of individual calls based on five temporal parameters. A high level of individual distinctiveness was also found in the fine temporal structure of calls. When information on fine structure was added to the discriminant function analysis, classification of calls to individuals was improved to 100%. The structure of an individual's call was found to be consistent over the 2-year period. This study highlighted the problemsassociated with setting criteria for re-identifying individuals by calls. The consequences of these problems and the potential of vocal individuality for providing monitoring information in this species are discussed.     

基于语谱图特征信息分割提取的声景观中鸟类生物多样性分析

声学手段是监测和研究生态系统生物活动规律、评价生态系统健康状况的一种新方法,声景观生态学也是景观生态学的一个新兴研究领域。声景指数是描述复杂的音频数据生态学特征的有效方法,但是,单一的声景指数并不能有效的指示物种的真实丰度。在充分挖掘音频文件时频结构特征的基础上,将遥感领域常用的面向对象图像分割技术引入语谱图分割,并提出了适合于自然界鸟类生物多样性提取的知识规则和斑块统计分析方法。研究实验在杭州植物园的不同区域布点采集音频数据,研究结果表明:鸟类的多样性与地物景观类型和人类活动影响程度密切相关,鸟类叫声的中心频率集中分布在2.5—4.5k Hz之间,最低频率分布在0.67—2.1k Hz之间,最高频率分布在7.6—8.9k Hz之间,人类活动较多的区域,鸟类活动较少且叫声更短促。此外,该方法提取的斑块面积周长比参数,可以定量的反映鸟类叫声的婉转程度。     

Syntactic structure and geographical dialects in the songs of male rock hyraxes

Few mammalian species produce vocalizations that are as richly structured as bird songs, and this greatly restricts the capacity for information transfer. Syntactically complex mammalian vocalizations have been previously studied only in primates, cetaceans and bats. We provide evidence of complex syntactic vocalizations in a small social mammal: the rock hyrax (Procavia capensis: Hyracoidea). We adopted three algorithms, commonly used in genetic sequence analysis and information theory, to examine the order of syllables in hyrax calls. Syntactic dialects exist, and the syntax of hyrax calls is significantly different between different regions in Israel. Call syntax difference is positively correlated to geographical distance over short distances. No correlation is found over long distances, which may reflect limited dispersal movement. These findings indicate that rich syntactic structure is more common in the vocalizations of mammalian taxa than previously thought and suggest the possibility of vocal production learning in the hyrax.     

Vocal activity rate index: a useful method to infer terrestrial bird abundance with acoustic monitoring

Autonomous recording units have been widely used in a large number of bird studies in recent years, but challenges remain in estimating abundance based on acoustic monitoring. We tested whether vocal activity rate index (VAR; the number of songs per unit time for a species), recorded using autonomous recording units, was related to population abundance in two terrestrial bird species, the European Bee‐eater Merops apiaster and the Dupont's Lark Chersophilus duponti. We took recordings at sites where censuses were also carried out to estimate local populations around recorders. We found a positive and significant relationship for the two monitored species. Although our results are not conclusive, the strong and significant relationship found for both monitored species suggests that VAR may be used to infer bird abundance around recorders in terrestrial species. We describe five logical steps for using the VAR with autonomous recording units in other species to guide future studies.     

The development of stimulus‐specific auditory responses requires song exposure in male but not female zebra finches

Juvenile male zebra finches develop their song by imitation. Females do not sing but are attracted to males' songs. With functional magnetic resonance imaging and event‐related potentials we tested how early auditory experience shapes responses in the auditory forebrain of the adult bird. Adult male birds kept in isolation over the sensitive period for song learning showed no consistency in auditory responses to conspecific songs, calls, and syllables. Thirty seconds of song playback each day over development, which is sufficient to induce song imitation, was also sufficient to shape stimulus‐specific responses. Strikingly, adult females kept in isolation over development showed responses similar to those of males that were exposed to songs. We suggest that early auditory experience with songs may be required to tune perception toward conspecific songs in males, whereas in females song selectivity develops even without prior exposure to song. © 2009 Wiley Periodicals, Inc. Develop Neurobiol 2010     

Testing acoustic versus physical marking: two complementary methods for individual-based monitoring of elusive species

Individuals of some species differentiate each other on the basis of the acoustic features of their vocalizations, and this can be used in individual‐based population monitoring studies. No research has tested for the effectiveness of individual marking through voice recognition as compared to traditional monitoring methods relying on physical marks. We compared voice recognition and physical marking using the Dupont's lark Chersophilus duponti as a study species. This bird needs to be attracted with playback in order to be seen (or captured). We first demonstrated that the territorial calls from a sample of banded males were individually distinctive and constant over time by means of discriminant function analysis, which correctly classified 100% of marked males. Then, we applied similarity techniques on call spectrotemporal features to define a threshold value of similarity within banded individuals, to be combined with qualitative spectrogram inspection for the classification of all recorded birds. Eventually, we compared the voice and the capture samples, to test for differences in relation to re‐location rate, territory fidelity and dispersal movements both within and between years. Voice recognition was less time‐consuming than capture‐recapture method in the field, but it was useless for monitoring yearlings in call development stage. The two methods provided the same results in terms of territory fidelity and dispersal movements, but differed in re‐location rates, which were significantly greater in the case of voice recognition method. By means of physical captures we possibly trapped a large sample of young and silent floaters, with low probability of recapture or recording. This mismatch between methods could bias the estimates of annual survival, which strongly depend on re‐location rates. We suggest considering the two methods as complementary rather than alternatives for monitoring populations. Each technique offers unique information, and the two sources should be combined to provide correction factors that would eventually sharpen our knowledge on bird population ecology.