Evolutionary novelties underlie sound production in baleen whales.

Baleen whales (mysticetes) use vocalizations to mediate their complex social and reproductive behaviours in vast, opaque marine environments1. Adapting to an obligate aquatic lifestyle demanded fundamental physiological changes to efficiently produce sound, including laryngeal specializations2-4. Whereas toothed whales (odontocetes) evolved a nasal vocal organ5, mysticetes have been thought to use the larynx for sound production1,6-8. However, there has been no direct demonstration that the mysticete larynx can phonate, or if it does, how it produces the great diversity of mysticete sounds9. Here we combine experiments on the excised larynx of three mysticete species with detailed anatomy and computational models to show that mysticetes evolved unique laryngeal structures for sound production. These structures allow some of the largest animals that ever lived to efficiently produce frequency-modulated, low-frequency calls. Furthermore, we show that this phonation mechanism is likely to be ancestral to all mysticetes and shares its fundamental physical basis with most terrestrial mammals, including humans10, birds11, and their closest relatives, odontocetes5. However, these laryngeal structures set insurmountable physiological limits to the frequency range and depth of their vocalizations, preventing them from escaping anthropogenic vessel noise12,13 and communicating at great depths14, thereby greatly reducing their active communication range.

Bats expand their vocal range by recruiting different laryngeal structures for echolocation and social communication.

Echolocating bats produce very diverse vocal signals for echolocation and social communication that span an impressive frequency range of 1 to 120 kHz or 7 octaves. This tremendous vocal range is unparalleled in mammalian sound production and thought to be produced by specialized laryngeal vocal membranes on top of vocal folds. However, their function in vocal production remains untested. By filming vocal membranes in excised bat larynges (Myotis daubentonii) in vitro with ultra-high-speed video (up to 250,000 fps) and using deep learning networks to extract their motion, we provide the first direct observations that vocal membranes exhibit flow-induced self-sustained vibrations to produce 10 to 95 kHz echolocation and social communication calls in bats. The vocal membranes achieve the highest fundamental frequencies (fo's) of any mammal, but their vocal range is with 3 to 4 octaves comparable to most mammals. We evaluate the currently outstanding hypotheses for vocal membrane function and propose that most laryngeal adaptations in echolocating bats result from selection for producing high-frequency, rapid echolocation calls to catch fast-moving prey. Furthermore, we show that bats extend their lower vocal range by recruiting their ventricular folds-as in death metal growls-that vibrate at distinctly lower frequencies of 1 to 5 kHz for producing agonistic social calls. The different selection pressures for echolocation and social communication facilitated the evolution of separate laryngeal structures that together vastly expanded the vocal range in bats.

Male and female syringeal muscles exhibit superfast shortening velocities in zebra finches.

Vocalisations play a key role in the communication behaviour of many vertebrates. Vocal production requires extremely precise motor control, which is executed by superfast vocal muscles that can operate at cycle frequencies over 100 Hz and up to 250 Hz. The mechanical performance of these muscles has been quantified with isometric performance and the workloop technique, but owing to methodological limitations we lack a key muscle property characterising muscle performance, the force-velocity relationship. Here, we quantified the force-velocity relationship in zebra finch superfast syringeal muscles using the isovelocity technique and tested whether the maximal shortening velocity is different between males and females. We show that syringeal muscles exhibit high maximal shortening velocities of 25L0 s-1 at 30°C. Using Q10-based extrapolation, we estimate they can reach 37-42L0 s-1 on average at body temperature, exceeding other vocal and non-avian skeletal muscles. The increased speed does not adequately compensate for reduced force, which results in low power output. This further highlights the importance of high-frequency operation in these muscles. Furthermore, we show that isometric properties positively correlate with maximal shortening velocities. Although male and female muscles differ in isometric force development rates, maximal shortening velocity is not sex dependent. We also show that cyclical methods to measure force-length properties used in laryngeal studies give the same result as conventional stepwise methodologies, suggesting either approach is appropriate. We argue that vocal behaviour may be affected by the high thermal dependence of superfast vocal muscle performance.

A flexible carbon nanotube electrode array for acute in vivo EMG recordings.

Executing complex behaviors requires precise control of muscle activity. Our understanding of how the nervous system learns and controls motor skills relies on recording electromyographic (EMG) signals from multiple muscles that are engaged in the motor task. Despite recent advances in tools for monitoring and manipulating neural activity, methods for recording in situ spiking activity in muscle fibers have changed little in recent decades. Here, we introduce a novel experimental approach to recording high-resolution EMG signals using parylene-coated carbon nanotube fibers (CNTFs). These fibers are fabricated via a wet spinning process and twisted together to create a bipolar electrode. Single CNTFs are strong, extremely flexible, small in diameter (14-24 µm), and have low interface impedance. We present two designs to build bipolar electrode arrays that, due to the small size of CNTF, lead to high spatial resolution EMG recordings. To test the EMG arrays, we recorded the activity of small (4 mm length) vocal muscles in songbirds in an acute setting. CNTF arrays were more flexible and yielded multiunit/bulk EMG recordings with higher SNR compared with stainless steel wire electrodes. Furthermore, we were able to record single-unit recordings not previously reported in these small muscles. CNTF electrodes are therefore well-suited for high-resolution EMG recording in acute settings, and we present both opportunities and challenges for their application in long-term chronic recordings.NEW & NOTEWORTHY We introduce a novel approach to record high-resolution EMG signals in small muscles using extremely strong and flexible carbon nanotube fibers (CNTFs). We test their functionality in songbird vocal muscles. Acute EMG recordings successfully yielded multiunit recordings with high SNR. Furthermore, they successfully isolated single-unit spike trains from CNTF recordings. CNTF electrodes have great potential for chronic EMG studies of small, deep muscles that demand high electrode flexibility and strength.

High-fidelity continuum modeling predicts avian voiced sound production.

Voiced sound production is the primary form of acoustic communication in terrestrial vertebrates, particularly birds and mammals, including humans. Developing a causal physics-based model that ultimately links descending vocal motor control to tissue vibration and sound requires embodied approaches that include realistic representations of voice physiology. Here, we first implement and then experimentally test a high-fidelity three-dimensional (3D) continuum model for voiced sound production in birds. Driven by individual-based physiologically quantifiable inputs, combined with noninvasive inverse methods for tissue material parameterization, our model accurately predicts observed key vibratory and acoustic performance traits. These results demonstrate that realistic models lead to accurate predictions and support the continuum model approach as a critical tool toward a causal model of voiced sound production.

Aerodynamics and motor control of ultrasonic vocalizations for social communication in mice and rats.

BACKGROUND: Rodent ultrasonic vocalizations (USVs) are crucial to their social communication and a widely used translational tool for linking gene mutations to behavior. To maximize the causal interpretation of experimental treatments, we need to understand how neural control affects USV production. However, both the aerodynamics of USV production and its neural control remain poorly understood. RESULTS: Here, we test three intralaryngeal whistle mechanisms-the wall and alar edge impingement, and shallow cavity tone-by combining in vitro larynx physiology and individual-based 3D airway reconstructions with fluid dynamics simulations. Our results show that in the mouse and rat larynx, USVs are produced by a glottal jet impinging on the thyroid inner wall. Furthermore, we implemented an empirically based motor control model that predicts motor gesture trajectories of USV call types. CONCLUSIONS: Our results identify wall impingement as the aerodynamic mechanism of USV production in rats and mice. Furthermore, our empirically based motor control model shows that both neural and anatomical components contribute to USV production, which suggests that changes in strain specific USVs or USV changes in disease models can result from both altered motor programs and laryngeal geometry. Our work provides a quantitative neuromechanical framework to evaluate the contributions of brain and body in shaping USVs and a first step in linking descending motor control to USV production.

Increasing Muscle Speed Drives Changes in the Neuromuscular Transform of Motor Commands during Postnatal Development in Songbirds.

Progressive changes in vocal behavior over the course of vocal imitation leaning are often attributed exclusively to developing neural circuits, but the effects of postnatal body changes remain unknown. In songbirds, the syrinx transforms song system motor commands into sound and exhibits changes during song learning. Here we test the hypothesis that the transformation from motor commands to force trajectories by syringeal muscles functionally changes over vocal development in zebra finches. Our data collected in both sexes show that, only in males, muscle speed significantly increases and that supralinear summation occurs and increases with muscle contraction speed. Furthermore, we show that previously reported submillisecond spike timing in the avian cortex can be resolved by superfast syringeal muscles and that the sensitivity to spike timing increases with speed. Because motor neuron and muscle properties are tightly linked, we make predictions on the boundaries of the yet unknown motor code that correspond well with cortical activity. Together, we show that syringeal muscles undergo essential transformations during song learning that drastically change how neural commands are translated into force profiles and thereby acoustic features. We propose that the song system motor code must compensate for these changes to achieve its acoustic targets. Our data thus support the hypothesis that the neuromuscular transformation changes over vocal development and emphasizes the need for an embodied view of song motor learning.SIGNIFICANCE STATEMENT Fine motor skill learning typically occurs in a postnatal period when the brain is learning to control a body that is changing dramatically due to growth and development. How the developing body influences motor code formation and vice versa remains largely unknown. Here we show that vocal muscles in songbirds undergo critical transformations during song learning that drastically change how neural commands are translated into force profiles and thereby acoustic features. We propose that the motor code must compensate for these changes to achieve its acoustic targets. Our data thus support the hypothesis that the neuromuscular transformation changes over vocal development and emphasizes the need for an embodied view of song motor learning.

Motor control by precisely timed spike patterns.

A fundamental problem in neuroscience is understanding how sequences of action potentials ("spikes") encode information about sensory signals and motor outputs. Although traditional theories assume that this information is conveyed by the total number of spikes fired within a specified time interval (spike rate), recent studies have shown that additional information is carried by the millisecond-scale timing patterns of action potentials (spike timing). However, it is unknown whether or how subtle differences in spike timing drive differences in perception or behavior, leaving it unclear whether the information in spike timing actually plays a role in brain function. By examining the activity of individual motor units (the muscle fibers innervated by a single motor neuron) and manipulating patterns of activation of these neurons, we provide both correlative and causal evidence that the nervous system uses millisecond-scale variations in the timing of spikes within multispike patterns to control a vertebrate behavior-namely, respiration in the Bengalese finch, a songbird. These findings suggest that a fundamental assumption of current theories of motor coding requires revision.

Quantifying syringeal dynamics in vitro using electroglottography.

The complex and elaborate vocalizations uttered by many of the 10,000 extant bird species are considered a major driver in their evolutionary success, warranting study of the underlying mechanisms of vocal production. Additionally, birdsong has developed into a highly productive model system for vocal imitation learning and motor control, where, in contrast to humans, we have experimental access to the entire neuromechanical control loop. In human voice production, complex laryngeal geometry, vocal fold tissue properties, airflow and laryngeal musculature all interact to ultimately control vocal fold kinematics. Quantifying vocal fold kinematics is thus critical to understanding neuromechanical control of voiced sound production, but in vivo imaging of vocal fold kinematics in birds is experimentally challenging. Here, we adapted and tested electroglottography (EGG) as a novel tool for examining vocal fold kinematics in the avian vocal organ, the syrinx. We furthermore imaged and quantified syringeal kinematics in the pigeon (Columba livia) syrinx with unprecedented detail. Our results show that EGG signals predict (1) the relative amount of contact between the avian equivalent of vocal folds and (2) essential parameters describing vibratory kinematics, such as fundamental frequency, and timing of syringeal opening and closing events. As such, EGG provides novel opportunities for measuring syringeal vibratory kinematic parameters in vivo Furthermore, the opportunity for imaging syringeal vibratory kinematics from multiple planar views (horizontal and coronal) simultaneously promotes birds as an excellent model system for studying kinematics and control of voiced sound production in general, including in humans and other mammals.

Plasticity in mitochondrial cristae density allows metabolic capacity modulation in human skeletal muscle.

KEY POINTS: In human skeletal muscles, the current view is that the capacity for mitochondrial energy production, and thus endurance capacity, is set by the mitochondria volume. However, increasing the mitochondrial inner membrane surface comprises an alternative mechanism for increasing the energy production capacity. In the present study, we show that mitochondrial inner membranes in leg muscles of endurance-trained athletes have an increased ratio of surface per mitochondrial volume. We show a positive correlation between this ratio and whole body oxygen uptake and muscle fibre mitochondrial content. The results obtained in the present study help us to understand modulation of mitochondrial function, as well as how mitochondria can increase their oxidative capacity with increased demand. ABSTRACT: Mitochondrial energy production involves the movement of protons down a large electrochemical gradient via ATP synthase located on the folded inner membrane, known as cristae. In mammalian skeletal muscle, the density of cristae in mitochondria is assumed to be constant. However, recent experimental studies have shown that respiration per mitochondria varies. Modelling studies have hypothesized that this variation in respiration per mitochondria depends on plasticity in cristae density, although current evidence for such a mechanism is lacking. In the present study, we confirm this hypothesis by showing that, in human skeletal muscle, and in contrast to the current view, the mitochondrial cristae density is not constant but, instead, exhibits plasticity with long-term endurance training. Furthermore, we show that frequently recruited mitochondria-enriched fibres have significantly increased cristae density and that, at the whole-body level, muscle mitochondrial cristae density is a better predictor of maximal oxygen uptake rate than muscle mitochondrial volume. Our findings establish an elevating mitochondrial cristae density as a regulatory mechanism for increasing metabolic power in human skeletal muscle. We propose that this mechanism allows evasion of the trade-off between cell occupancy by mitochondria and other cellular constituents, as well as improved metabolic capacity and fuel catabolism during prolonged elevated energy requirements.

Multifunctional and Context-Dependent Control of Vocal Acoustics by Individual Muscles.

The relationship between muscle activity and behavioral output determines how the brain controls and modifies complex skills. In vocal control, ensembles of muscles are used to precisely tune single acoustic parameters such as fundamental frequency and sound amplitude. If individual vocal muscles were dedicated to the control of single parameters, then the brain could control each parameter independently by modulating the appropriate muscle or muscles. Alternatively, if each muscle influenced multiple parameters, a more complex control strategy would be required to selectively modulate a single parameter. Additionally, it is unknown whether the function of single muscles is fixed or varies across different vocal gestures. A fixed relationship would allow the brain to use the same changes in muscle activation to, for example, increase the fundamental frequency of different vocal gestures, whereas a context-dependent scheme would require the brain to calculate different motor modifications in each case. We tested the hypothesis that single muscles control multiple acoustic parameters and that the function of single muscles varies across gestures using three complementary approaches. First, we recorded electromyographic data from vocal muscles in singing Bengalese finches. Second, we electrically perturbed the activity of single muscles during song. Third, we developed an ex vivo technique to analyze the biomechanical and acoustic consequences of single-muscle perturbations. We found that single muscles drive changes in multiple parameters and that the function of single muscles differs across vocal gestures, suggesting that the brain uses a complex, gesture-dependent control scheme to regulate vocal output.

Vocal production complexity correlates with neural instructions in the oyster toadfish (Opsanus tau).

Sound communication is fundamental to many social interactions and essential to courtship and agonistic behaviours in many vertebrates. The swimbladder and associated muscles in batrachoidid fishes (midshipman and toadfish) is a unique vertebrate sound production system, wherein fundamental frequencies are determined directly by the firing rate of a vocal-acoustic neural network that drives the contraction frequency of superfast swimbladder muscles. The oyster toadfish boatwhistle call starts with an irregular sound waveform that could be an emergent property of the peripheral nonlinear sound-producing system or reflect complex encoding in the central nervous system. Here, we demonstrate that the start of the boatwhistle is indicative of a chaotic strange attractor, and tested whether its origin lies in the peripheral sound-producing system or in the vocal motor network. We recorded sound and swimbladder muscle activity in awake, freely behaving toadfish during motor nerve stimulation, and recorded sound, motor nerve and muscle activity during spontaneous grunts. The results show that rhythmic motor volleys do not cause complex sound signals. However, arrhythmic recruitment of swimbladder muscle during spontaneous grunts correlates with complex sounds. This supports the hypothesis that the irregular start of the boatwhistle is encoded in the vocal pre-motor neural network, and not caused by peripheral interactions with the sound-producing system. We suggest that sound production system demands across vocal tetrapods have selected for muscles and motorneurons adapted for speed, which can execute complex neural instructions into equivalently complex vocalisations.

The songbird syrinx morphome: a three-dimensional, high-resolution, interactive morphological map of the zebra finch vocal organ.

BACKGROUND: Like human infants, songbirds learn their species-specific vocalizations through imitation learning. The birdsong system has emerged as a widely used experimental animal model for understanding the underlying neural mechanisms responsible for vocal production learning. However, how neural impulses are translated into the precise motor behavior of the complex vocal organ (syrinx) to create song is poorly understood. First and foremost, we lack a detailed understanding of syringeal morphology. RESULTS: To fill this gap we combined non-invasive (high-field magnetic resonance imaging and micro-computed tomography) and invasive techniques (histology and micro-dissection) to construct the annotated high-resolution three-dimensional dataset, or morphome, of the zebra finch (Taeniopygia guttata) syrinx. We identified and annotated syringeal cartilage, bone and musculature in situ in unprecedented detail. We provide interactive three-dimensional models that greatly improve the communication of complex morphological data and our understanding of syringeal function in general. CONCLUSIONS: Our results show that the syringeal skeleton is optimized for low weight driven by physiological constraints on song production. The present refinement of muscle organization and identity elucidates how apposed muscles actuate different syringeal elements. Our dataset allows for more precise predictions about muscle co-activation and synergies and has important implications for muscle activity and stimulation experiments. We also demonstrate how the syrinx can be stabilized during song to reduce mechanical noise and, as such, enhance repetitive execution of stereotypic motor patterns. In addition, we identify a cartilaginous structure suited to play a crucial role in the uncoupling of sound frequency and amplitude control, which permits a novel explanation of the evolutionary success of songbirds.

Biomechanics of sound production in high-pitched classical singing.

Voice production of humans and most mammals is governed by the MyoElastic-AeroDynamic (MEAD) principle, where an air stream is modulated by self-sustained vocal fold oscillation to generate audible air pressure fluctuations. An alternative mechanism is found in ultrasonic vocalizations of rodents, which are established by an aeroacoustic (AA) phenomenon without vibration of laryngeal tissue. Previously, some authors argued that high-pitched human vocalization is also produced by the AA principle. Here, we investigate the so-called "whistle register" voice production in nine professional female operatic sopranos singing a scale from C6 (≈ 1047 Hz) to G6 (≈ 1568 Hz). Super-high-speed videolaryngoscopy revealed vocal fold collision in all participants, with closed quotients from 30 to 73%. Computational modeling showed that the biomechanical requirements to produce such high-pitched voice would be an increased contraction of the cricothyroid muscle, vocal fold strain of about 50%, and high subglottal pressure. Our data suggest that high-pitched operatic soprano singing uses the MEAD mechanism. Consequently, the commonly used term "whistle register" does not reflect the physical principle of a whistle with regard to voice generation in high pitched classical singing.

How the bat got its buzz.

Since the discovery of echolocation in bats, the final phase of an attack on a flying insect, the 'terminal buzz', has proved enigmatic. During the buzz, bats increase information update rates by producing vocalizations up to 220 times s(-1). The buzz's ubiquity in hawking and trawling bats implies its importance for hunting success. Superfast muscles, previously unknown in mammals, are responsible for the extreme vocalization rate. Some bats produce a second phase-buzz II-defined by a large drop in the fundamental frequency (F(0)) of their calls. By doing so, bats broaden their acoustic field of view and should thereby reduce the likelihood of insect escape. We make the case that the buzz was a critical adaptation for capturing night-flying insects, and suggest that the drop in F(0) during buzz II requires novel, unidentified laryngeal mechanisms in order to counteract increasing muscle tension. Furthermore, we propose that buzz II represents a countermeasure against the evasive flight of eared prey in the evolutionary arms-race that saw the independent evolution of bat-detecting ears in various groups of night-flying insects.

Toothed whales use distinct vocal registers for echolocation and communication.

Echolocating toothed whales (odontocetes) capture fast-moving prey in dark marine environments, which critically depends on their ability to generate powerful, ultrasonic clicks. How their supposedly air-driven sound source can produce biosonar clicks at depths of >1000 meters, while also producing rich vocal repertoires to mediate complex social communication, remains unknown. We show that odontocetes possess a sound production system based on air driven through nasal passages that is functionally analogous to laryngeal and syringeal sound production. Tissue vibration in different registers produces distinct echolocation and communication signals across all major odontocete clades, and thus provides a physiological basis for classifying their vocal repertoires. The vocal fry register is used by species from porpoises to sperm whales for generating powerful, highly air-efficient echolocation clicks.

Dynamic System Coupling in Voice Production.

Voice is a major means of communication for humans, non-human mammals and many other vertebrates like birds and anurans. The physical and physiological principles of voice production are described by two theories: the MyoElastic-AeroDynamic (MEAD) theory and the Source-Filter Theory (SFT). While MEAD employs a multiphysics approach to understand the motor control and dynamics of self-sustained vibration of vocal folds or analogous tissues, SFT predominantly uses acoustics to understand spectral changes of the source via linear propagation through the vocal tract. Because the two theories focus on different aspects of voice production, they are often applied distinctly in specific areas of science and engineering. Here, we argue that the MEAD and the SFT are linked integral aspects of a holistic theory of voice production, describing a dynamically coupled system. The aim of this manuscript is to provide a comprehensive review of both the MEAD and the source-filter theory with its nonlinear extension, the latter of which suggests a number of conceptual similarities to sound production in brass instruments. We discuss the application of both theories to voice production of humans as well as of animals. An appraisal of voice production in the light of non-linear dynamics supports the notion that voice production can best be described with a systems view, considering coupled systems rather than isolated contributions of individual sub-systems.

Daily vocal exercise is necessary for peak performance singing in a songbird.

Vocal signals, including human speech and birdsong, are produced by complicated, precisely coordinated body movements, whose execution is fitness-determining in resource competition and mate choice. While the acquisition and maintenance of motor skills generally requires practice to develop and maintain both motor circuitry and muscle performance, it is unknown whether vocal muscles, like limb muscles, exhibit exercise-induced plasticity. Here, we show that juvenile and adult zebra finches (Taeniopygia castanotis) require daily vocal exercise to first gain and subsequently maintain peak vocal muscle performance. Experimentally preventing male birds from singing alters both vocal muscle physiology and vocal performance within days. Furthermore, we find females prefer song of vocally exercised males in choice experiments. Vocal output thus contains information on recent exercise status, and acts as an honest indicator of past exercise investment in songbirds, and possibly in all vocalising vertebrates.

Peak performance singing requires daily vocal exercise in songbirds.

Vocal signals mediate much of human and non-human communication. Key performance traits - such as repertoire size, speed and accuracy of delivery - affect communication efficacy in fitness-decisive contexts such as mate choice and resource competition 1 . Specialized fast vocal muscles 2,3 are central to accurate sound production 4 , but it is unknown whether vocal, like limb muscles 5,6 , need exercise to gain and maintain peak performance 7,8 . Here, we show that for song development in juvenile songbirds, the closest analogue to human speech acquisition 9 , regular vocal muscle exercise is crucial to achieve adult peak muscle performance. Furthermore, adult vocal muscle performance reduces within two days of abolishing exercise, leading to downregulation of critical proteins transforming fast to slower muscle fibre types. Daily vocal exercise is thus required to both gain and maintain peak vocal muscle performance, and if absent changes vocal output. We show that conspecifics can detect these acoustic changes and females prefer the song of exercised males. Song thus contains information on recent exercise status of the sender. Daily investment in vocal exercise to maintain peak performance is an unrecognized cost of singing and could explain why many birds sing daily even under adverse conditions 10 . Because neural regulation of syringeal and laryngeal muscle plasticity is equivalent, vocal output may reflect recent exercise status in all vocalizing vertebrates.

Driving singing behaviour in songbirds using a multi-modal, multi-agent virtual environment.

Interactive biorobotics provides unique experimental potential to study the mechanisms underlying social communication but is limited by our ability to build expressive robots that exhibit the complex behaviours of birds and small mammals. An alternative to physical robots is to use virtual environments. Here, we designed and built a modular, audio-visual 2D virtual environment that allows multi-modal, multi-agent interaction to study mechanisms underlying social communication. The strength of the system is an implementation based on event processing that allows for complex computation. We tested this system in songbirds, which provide an exceptionally powerful and tractable model system to study social communication. We show that pair-bonded zebra finches (Taeniopygia guttata) communicating through the virtual environment exhibit normal call timing behaviour, males sing female directed song and both males and females display high-intensity courtship behaviours to their mates. These results suggest that the environment provided is sufficiently natural to elicit these behavioral responses. Furthermore, as an example of complex behavioral annotation, we developed a fully unsupervised song motif detector and used it to manipulate the virtual social environment of male zebra finches based on the number of motifs sung. Our virtual environment represents a first step in real-time automatic behaviour annotation and animal-computer interaction using higher level behaviours such as song. Our unsupervised acoustic analysis eliminates the need for annotated training data thus reducing labour investment and experimenter bias.