Fidelity of vocal mimicry: identification and accuracy of mimicry of heterospecific alarm calls by the brown thornbill

Avian vocal mimicry has been studied for decades, but little is known about its function or requirements for accurate imitation. Furthermore, progress is hampered by the difficulty in identifying which vocalizations are indeed mimetic. We tested historical claims of vocal mimicry in the brown thornbill, Acanthiza pusilla, using a combination of human and computer methods to identify mimicry, followed by comparisons of acoustic similarity with model vocalizations. We recorded vocalizations of brown thornbills and sympatric heterospecifics while undisturbed and during mist net capture or the presence of natural or model predators. We then cross-validated human classification of mimicry with computer classification based on spectrographic measurements and spectral cross-correlation. Finally, we quantified the accuracy of the most common imitations. Brown thornbills predominantly imitated alarm calls given by heterospecifics towards aerial predators, which function in these models to provoke immediate flight by receivers. Human and computer-based methods produced consistent results when identifying and classifying mimicry. Mimicked aerial alarms were not perfect imitations of their corresponding model alarms, but did retain specific acoustic properties previously shown to be important for provoking immediate alarm responses. Although less accurate mimicry may reflect physiological constraints, we suggest that mimetic function, perhaps startling predators, only requires mimicry to retain features of model alarms that provoke immediate alarm responses by receivers. Understanding what factors influence the acoustic structure of mimetic vocalizations is essential in understanding the evolution of vocal mimicry, particularly with accumulating evidence that mimetic function does not always require perfect resemblance in other sensory modalities.