Measures of eastern quoll vocalisation extracted using PRAAT

Defining an acoustic repertoire is essential to understanding vocal signalling and communicative interactions within a species. Currently, quantitative and statistical definition is lacking for the vocalisations of many dasyurids, an important group of small to medium-sized marsupials from Australasia that includes the eastern quoll (Dasyurus viverrinus), a species of conservation concern. Beyond generating a better understanding of this species' social interactions, determining an acoustic repertoire will further improve detection rates and inference of vocalisations gathered by automated bioacoustic recorders. Hence, this study investigated eastern quoll vocalisations using objective signal processing techniques to quantitatively analyse spectrograms recorded from 15 different individuals. Recordings were collected from Secret Creek Sanctuary in Lithgow in conjunction with observations of the behaviours associated with each vocalisation to develop an acoustic-based behavioural repertoire for the species. Vocalisation measures were extracted using narrowband spectrograms (FFT method, window length 0.05 sec, dynamic range = 70 dB, time-steps = 1,000, frequency steps = 250, Gaussian window shape) produced in the program PRAAT (5.3.84 DSP Package). Source-related parameters using an autocorrelation method were used to detect the fundamental frequency (F0) contour from which measures of Duration, Median F0, Mean F0, Minimum F0, Maximum F0, Range of F0, Standard deviation of F0, Noise-to-Harmonics ratio, Jitter and Shimmer were extracted. Additionally intensity contours were extracted for each call to measure the Minimum amplitude, Maximum amplitude and Amplitude variation. Analysis of recordings produced a putative classification of five vocalisation types: Bark, Growl, Hiss, Cp-cp, and Chuck. These were most frequently observed during agonistic encounters between conspecifics, most likely as a graded sequence from Hisses occurring in a warning context through to Growls and finally Barks being given prior to, or during, physical confrontations between individuals. Quantitative and statistical methods were used to objectively establish the accuracy of these five putative call types. A multinomial logistic regression indicated a 97.27% correlation with the perceptual classification, demonstrating support for the five different vocalisation types. This putative classification was further supported by hierarchical cluster analysis and silhouette information that determined the optimal number of clusters to be five. Minor disparity between the objective and perceptual classifications was potentially the result of gradation between vocalisations, or subtle differences present within vocalisations not discernible to the human ear. The implication of these different vocalisations and their given context is discussed in relation to the ecology of the species and the potential application of passive acoustic monitoring techniques.

界定物种的发声库（acoustic repertoire），是理解该物种种内声学信号传递与交流互动的核心前提。目前，包括受保护关注物种东部袋鼬（Dasyurus viverrinus）在内的澳大拉西亚中小型有袋类重要类群——袋鼬科（Dasyuridae）动物的多数发声行为，尚未形成定量与统计层面的明确定义。明确该物种的发声库，不仅能深化对其社会互动模式的认知，还可提升自动化生物声学记录仪所采集发声信号的识别率与推断精度。
为此，本研究采用客观信号处理技术，对15只不同个体的东部袋鼬发声语谱图（spectrogram）开展定量分析。研究录音采集自利斯戈市的秘密溪保护区（Secret Creek Sanctuary），同时同步记录了对应每种发声行为的行为场景，以此构建该物种的声学行为发声库。
本研究使用PRAAT（5.3.84 DSP软件包）生成窄带语谱图（快速傅里叶变换（FFT）方法，窗长0.05秒，动态范围70dB，时间步长1000，频率步长250，高斯窗型），以此提取发声相关特征参数。研究采用自相关法提取声源相关参数，以检测基频（F0）轮廓，并由此提取以下特征：时长、基频中位数、基频均值、基频最小值、基频最大值、基频极差、基频标准差、噪声谐波比、基频微扰（Jitter）与振幅微扰（Shimmer）。此外，本研究还为每一类发声信号提取强度轮廓，以计算最小振幅、最大振幅与振幅变异度。
通过对录音的分析，本研究初步推定该物种存在5类发声类型：吠叫（Bark）、低吼（Growl）、嘶鸣（Hiss）、嗒嗒声（Cp-cp）与啾鸣（Chuck）。上述发声类型多出现于同种个体的对抗性互动中，且大概率呈现出梯度序列：从用于警戒场景的嘶鸣，过渡到低吼，最终在个体间肢体冲突发生前或冲突进行时发出吠叫。
本研究采用定量与统计方法，客观验证了这5类推定发声类型的准确性：多项逻辑回归（multinomial logistic regression）分析显示，其与人工感知分类的契合度达97.27%，验证了该5类发声类型的合理性。层次聚类分析（hierarchical cluster analysis）与轮廓系数（silhouette）分析进一步佐证了该推定分类的合理性，二者均确定最优聚类数为5。客观分类与人工感知分类间的细微偏差，可能源于发声类型间的梯度过渡，或是发声内部存在人类听觉难以辨识的细微差异。
本研究还结合该物种的生态学特征，探讨了不同发声类型及其对应场景的生态学内涵，以及被动声学监测技术的潜在应用价值。