Imitation of human music by non-human animals

Some non-human animals, such as songbirds, can add novel acoustic patterns to their own vocal repertoires by listening to sounds [5, 6]; they are considered vocal learners. However, in most vocal learning animals, the innate constraints for incorporating novel sounds is much stricter than in humans. For example, juvenile finches learn songs from their father very well; however, they copy songs imperfectly when the tutor is a foster father of a different species [7, 8]. Therefore, it makes sense that most animals do not imitate melodies of human music.

Some vocal learning animals imitate human speech and/or artificial sounds. The African grey parrot, Alex [9], and the harbor seal, Hoover [10], may be the most famous examples. More recently, a study reported that an elephant produced human speech sounds [11]. Some vocal learning animals also imitate various artificial sounds. Lyrebirds imitate the noises of a chainsaw [12], for example. However, human music has unique qualities apart from speech and other artificial sounds; music has harmonic syntax, rhythmic syntax, and is characterized by meter, grouping and hierarchy. Some of these qualities are shared in common with language, but not all of them [13]. Thus, the cognitive processes involved in musicality should differ from those used in the perception and production of other types of sound sequences, including human speech. Moreover, it is plausible that the acoustic structure of human music is quite different from the acoustic structure of natural animal vocalizations. Thus, musical melodies may be suitable imitation models for evaluating the flexibility of cognitive processing involved in vocal learning in animals. Therefore, investigating whether animals can imitate human music may be an interesting research theme, and may provide more insight for further understanding of their cognitive capabilities, beyond whether they can imitate human speech and/or artificial noises.

Anecdotally, many videos have been uploaded to online databases (such as YouTube) by aviculturists, in which cockatiels imitate melodies of human music with their whistle-like vocal sequences. Examples include the “Mickey Mouse Club March” and the theme song of “My Neighbor Totoro”. However, there are only a few academic publications describing imitation of human music by animals. One study showed that bullfinches imitated human music and sang the melody alternately with whistling produced by an experimenter [14]. Further, it was documented that a European starling named Kuro sang some melodies of human music [15]. Another study reported that gray seals learned to modify their vocalizations to match the frequency patterns of musical melodies that were composed of human vowel-like sounds under an operant conditioning paradigm [16] (note: in these studies, animals did not sing in unison). In sum, further academic documentation of music production by animals would be valuable for comparative approaches in relevant research fields, such as neuroscience, cognitive studies, and psychology.

Singing in unison by non-human animals

Many animals, including non-vocal learners and even invertebrates, sing songs together, forming duets or choruses [17]. As an example, howling by wolves results in acoustic patterns that form a heterophony [3]. As another example, two gibbons generate great calls together with the vocal timing of one individual likely depending on that of the other [18–20]. In addition, many studies documented songbirds singing together [21]. Songbirds are vocal learners; thus, singing together in songbirds is much more relevant to the present topic than similar behaviors in non-vocal learners. However, studies demonstrating singing in unison or, monophony (as opposed to simply singing at the same time) by non-human animals are quite rare, including those in songbirds. Likely, this is one reason why one author stated that “human beings are the only species to have evolved the ability to sing in unison in both the dimensions of rhythmic co-ordination and precise pitch attunement” [22]. As a matter of fact, it is known that a few avian species do sing in unison. Male and female mates of the forest weaver sing in unison, which were likely derived from a territorial display [23, 24]. White-browed sparrow weavers also sing in unison [25]. Lastly, male and female plain-tailed wrens sing duets antiphonally, but occasionally males and females sing songs in unison with near perfect synchrony [26–28]. In mammals, recently, a study reported that two male dolphins occasionally produced isochronous pop sound sequences in synchrony [29], which is likely the only academic report of non-human mammals vocalizing in unison thus far. However, this behavior is much simpler than the singing in unison performed by humans.

In sum, the examples of singing in unison by non-human animals reported in these studies were (i) only between conspecifics, (ii) occurred in specific contexts involving reproduction (including territorial defense), and (iii) used only vocal variations observed in the wild (the vocal system is optimized to use those sounds, making the behavior relatively easy; on the contrary, learning and producing a melody of human music may require much larger costs for the vocal system as described above). Therefore, the present study would provide further knowledge to understand the similarities and differences seen between humans and non-human animals with regard to musicality.

Introduction for the experiment

A research project was launched to investigate musicality in non-human animals. Cockatiels were chosen as the subjects because it is well known that they sing melodies of human music as described above. In the project, three hand-raised cockatiels learned to sing a marching sound: whistling of a melody similar to the “Mickey Mouse Club March” produced by a human (the experimenter). The melody was composed of the two parts; the first half (consisting of 11 notes) and the second half (consisting of 11 notes) separated by a long pause (640 ms, see Materials and Methods). Then, two of the birds (bird C and bird PY) spontaneously sang in unison with the whistling (see S1 Movie). Therefore, the following two experiments (with some predictions) were carried out to examine whether the birds would actively synchronize their vocalizations with a playback of a recording of the melody.

Experiment 1: A playback sequence was presented while a bird was singing

Using the updated recording system (see Materials and Methods), 7 recordings were obtained from bird C (C#1 − C#7 in Supplementary Information; hereafter, S1 File) and 5 recordings were obtained from bird PY (PY#1 − PY#5 in S1 File), in which the birds kept singing after the playback sounds were presented. The birds changed the pause duration of their own song (consistent with Prediction I). The actual onset of the second half of singing was delayed from the putative onset (the original playback sound was used as a standard reference to measure the delay; see Fig 1A upper and S1 File). The delay in singing (Y; indicated by green shading in Fig 1A) depended on the latency of the playback (X; indicated by yellow shading in Fig 1A). The duration of Y was longer when X was longer (Fig 1B). There was a significant correlation between X and Y (r = 0.972 [95% CI = 0.899 − 0.992], t = 12.975, df = 10, p < 0.001; individual results are also shown in S1 File). These results indicate that the birds actively adjusted their vocal timing to the playback sound.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Effects of playback sounds on ongoing singing.

These recordings were obtained from bird C (the same recordings are shown as C#3, C#4 and C#6 in S1 File). Model sounds (PB: playback, pink shading) were played back after the bird began singing (Vo.: blue shading). Yellow shading indicates the latency of the playback sound (X). Green shading indicates the duration between the onset of the second half of the standard reference (see main text) and the actual onset of the second half of singing (Y). Note that the onset of the second half of the playback sound and the singing started at nearly the same time (A). Correlation between the length of X and the length of Y is shown in Fig 1A. Each dot denotes values of X and Y obtained from a recording that includes bird singing and playback of the melody. When the length of X is longer, the length of Y is longer (B).

https://doi.org/10.1371/journal.pone.0256613.g001

Experiment 2: A bird began to sing following the presentation of a playback sequence

Bird C did not begin to sing following the presentation of a playback sequence. However, 15 recordings were obtained from bird PY, in which the bird sang songs following the presentation of playback sounds (PY#6 − PY#20 in S1 File). In 1 out of the 15 recordings, the bird sang the normal sequence of the melody with some irregular note intervals, as if he was trying to ignore the playback sound; then, he stopped singing in the middle of the second half (PY#18 in S1 File). In 2/15 recordings, the bird sang a limited part of the first half and then stopped singing suddenly (PY#19 and PY#20 in S1 File). However, in 9/15 recordings, the bird started singing and synchronized his vocal timing with the playback, either from the beginning of the second half by skipping the entire first half (5/15 recordings; Fig 2A; PY#6 − PY#10 in S1 File), or from the middle of the first half by skipping several (between 1 and 6) initial notes (4/15 recordings; Fig 2B; PY#11 − PY#14 in S1 File). Both of these adjustments are consistent with Prediction II. The bird occasionally sang songs spontaneously during the experimental period, so 46 song recordings were obtained without the playback sound. In these recordings, the bird always began singing from the beginning of the first half and never started in the middle of the melody (0/46 recordings), indicating that any abnormal singing observed was not caused by chance, but was in response to the playback sound (p < 0.001, Fisher’s exact test).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Examples of singing in response to playback sounds.

These recordings were obtained from bird PY. Starting from the second half (A; the same recordings are shown as PY#6 and PY#10 in S1 File). Starting from in the middle of the first half (B; the same recordings are shown as PY#12 and PY#13 in S1 File).

https://doi.org/10.1371/journal.pone.0256613.g002

In 5/15 recordings, bird PY started singing from the beginning of the sequence with a slight delay following a playback sound (PY#15 − PY#19 in S1 File). Each time, the latency was roughly the same value (mean = 243.9 ms, SD = 20.99 ms). Therefore, it is possible that these 5 songs were triggered by the first note of the playback sound. While the bird stopped singing in the middle of the normal song in 2 out of the 5 recordings (PY#18 and PY#19 in S1 File), he sang the song to the end in 3/5 recordings (Fig 3; PY#15, PY#16 and PY#17 in S1 File). However, in these 3 recordings, the last two notes of the first half were dropped and the songs were resumed at the onset of the second half, which did not happen when the bird was singing without the playback (0/46). Thus, it is likely that this behavior did not simply occur by chance (p = 0.029, Fisher’s exact test), suggesting the bird actively skipped those two notes to synchronize his vocal timing with the playback at the beginning of the second half. This adjustment is consistent with Prediction III.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Examples of irregular vocal sequences with playback of the melody.

These recordings were obtained from bird PY (the same recordings are shown as PY#15, PY#16 and PY#17 in S1 File). The vocalizations began following the playback (indicated by black arrows) and the last two notes of the first half were dropped (indicated by red circles).

https://doi.org/10.1371/journal.pone.0256613.g003

In Experiment 2, there were several examples in which Bird PY began to sing following the playback and it is possible that the bird entrained to the ongoing melody. If this is the case, then for the 12 recordings which were consistent with our predictions (9 songs for Prediction II and 3 songs for Prediction III) we might see a negative mean asynchrony (NMA). It is well known that when humans create a series of motor outputs synchronizing with a rhythmic stimulus, their motor outputs tend to slightly precede each stimulus presentation, which is termed NMA [30]. Also, NMA is considered one indication that animals have entrained to a rhythmic sequence [31]. To examine this possibility, the timing of the vocalization at the onset of the second half was compared to the timing of the playback at the onset of the second half. The results showed that singing started slightly but significantly faster than the playback (mean -135 ms, median -94 ms; V = 8, p = 0.012, Wilcoxon signed rank test [the null hypothesis; mu = 0]), which can be interpreted as an NMA. This result also supports the interpretation that the bird actively synchronized his vocal timing with the playback.

Why do cockatiels sing in unison?

To obtain an answer to this question, we can turn to previous studies investigating the vocal behavior of wild parrots, but there are only a few studies. Duets between conspecifics (sung in antiphony, not in unison) have been reported in wild Grey-headed parrots [32] and Yellow-naped amazons [33]. The duets of Yellow-naped amazons are produced by a male and female pair, which are sung for joint territory defense [34]. However, the behavioral contexts in which these wild parrots sing duets differ from the contexts for singing in unison by the cockatiels in this study.

As some authors described, studying the vocal learning of parrots in the wild is difficult due to several reasons: they stay high up in the canopy, move in and out of the foliage, and do not generally stay in one place for long periods of time [35, 36]. Thus, there are also currently no academic reports of vocal learning in wild cockatiels, to my knowledge. Moreover, there are few laboratory studies examining the vocal behavior of cockatiels, most of which describe only their short calls, and not songs [37–41]. As one exception, an author described that captive cockatiels often produced complex, long-lasting vocalizations which might appear to qualify as song [42]; however, the author did not describe the acoustic details or the function of these vocalizations. It is known among aviculturists that captive cockatiels often imitate human speech, music and other sounds; however, it is unclear whether cockatiels sing songs in the wild or not (and consequently, even if they do, the function is unknown). To the contrary, budgerigars very often sing warble songs. It is assumed that the songs are used for bonding among social companions, individual / group recognition, and as a badge of group membership [43]. Therefore, the speculation that cockatiels sing in unison for social bonding may be valid. Anecdotally, the cockatiels in this study often sang the musical sounds or imitated human speech when caregivers were leaving the aviary. Therefore, they may have done so to attract attention from humans, though there is no quantitative data to support this observation. This is consistent with the music and social bonding hypothesis [2]. This idea is also supported by the fact that no food reinforcements were necessary for the birds to sing in unison. We may think about another related question. The bullfinches [14] and European starling Kuro [15] described in the Introduction were hand-reared, as were the cockatiels in the present study. Therefore, these birds might recognize their human caregivers as their own conspecifics, and might recognize human music as vocalizations produced by conspecifics. As a result, they might acquire the melodies in the same way as many songbirds acquire songs from conspecifics. On the contrary, it is also possible that the birds learned the melody even though they recognized their human caregivers as heterospecfic animals and considered the melodies to be heterospecific vocalizations. However, there is currently a lack of quantitative data enabling further discussion of these possibilities. We must wait until the points described above have been tested experimentally.

How do cockatiels sing in unison?

Auditory-vocal mirroring in the neural circuit involved in vocal production and vocal learning in songbirds may partially explain the substrates for this behavior. Some neurons in the nucleus HVC (proper name, not an abbreviation) are activated not only when a bird produces a particular song note but also when the bird listens to the same song note [44]. Parrots have a neural circuit for vocal production and vocal learning similar to songbirds that contains the nucleus NLC (central nucleus of the lateral nidopallium) which corresponds to the songbird HVC [45]. Therefore, it is valid to speculate that some neurons involved in the production of the melody were also activated when birds listened to the playback of the melody as if the cockatiel was singing. The neural activity elicited from listening might lead to singing of the melody. Thus, the present findings may be a behavioral demonstration suggesting parrots have a neural system for auditory-vocal mirroring similar to songbirds. Another recent study recorded neural activity of HVC neurons in songbird pairs during duet singing. The authors reported that the degree of interindividual synchronization of neural activity was positively correlated with the degree of interindividual synchronization of vocal activity during alternating parts of duet bouts [46]. Therefore, the activity of NLC in cockatiels might be involved in the timing control of singing in unison.

Further, parrots have additional structures in their vocal learning nervous system which may be involved in this behavior. The parrot vocal nuclei have core regions and shell regions. Connections between the cores and shells are sparse within and among each vocal nucleus. Cores tend to project to cores, and shells tend to project to shells. This results in two parallel systems: a core system and a shell system [47]. The neural connectivity of the core vocal nuclei is similar to that of the song nuclei of songbirds, whereas the shell system is unique to parrots. The unique system may allow for more complex vocal communication abilities and greater auditory–motor entrainment than in other birds [48].

In addition, parrots may have excellent neural and psychological substrates to synchronize their motor outputs to external rhythmic stimuli. The cockatoo, Snowball, spontaneously showed entrainment through dance to various beats of human music [49], which was considered the first demonstration of this type of behavior by non-human animals. Other studies trained budgerigars to synchronize to metronomic sounds using operant conditioning techniques [31, 50]. While the impact of these budgerigar studies are less than that of the Snowball study, the results support the idea that some parrot species have an excellent capability for entrainment. Rhythmic entrainment should be one of the important factors required for singing in unison; thus, the neural and psychological substrates for rhythmic synchronization of body movements may have a linkage to that for rhythmic synchronization of vocalizations, at least in parrots.

In addition, some recent studies suggest that parrots may have some further capabilities involved in musicality. Snowball originally created diverse motor patterns for dancing that were performed with musical melodies [51]. A study reported that an African grey parrot was trained to produce a vocal sequence following a sound sequence produced from a piano. Each sequence produced by the bird followed musical rules similar to the previous piano sequence in terms of the frequency ratios between the notes [52]. Another study showed that wild palm cockatoos can make drumming sequences with regular intervals using several tools [53]. Moreover, a recent study reported that cockatiels manipulated several objects producing sounds [54]. The substrates for singing human music in unison by cockatiels may also be relevant to these behaviors.

Subjects

Male cockatiels were obtained from a local breeder at the age of approximately 25 days post-hatch (dph) and were kept in independent cages (370 mm (W)× 415 mm (D) × 440 mm (H)) placed in an aviary (25°C, 12:12h photoperiod) at Aichi University. The birds were hand-raised by human caregivers to facilitate the ability to mimic sounds produced by humans and frequently heard human voices; however, the birds were isolated from human music except the model melody as much as possible (they might be occasionally exposed to some sounds, such as chimes and sirens, when they were outside their rearing room).

Apparatus

Stimulus sounds (human whistles) and birds’ vocalizations were captured with a cardioid microphone (PRO35, Audio-technica, Japan) and recorded with a PCM recorder (DR-40, Teac, Japan). Stimulus sounds were presented through a loudspeaker (either AT-SP151 or AT-SP120, Audio-technica) during the playback experiment.

Stimulus

A typical series of whistle sounds (which was similar to a portion of the “Mickey Mouse Club March”; 8.1s duration, 107 bpm [beats per minute], frequency range: 1100−1950 Hz) produced by the experimenter was recorded as a Windows PCM file (.wav format; 44.1 kHz sampling rate) and band-pass filtered (500–4000 Hz) using software (SASLab Pro., Avisoft Bioacoustics, Germany). The sound sequence was used as a template (or, a model) for the imitation by the birds. The melody was composed of 22 notes and was divided into two halves (11 notes each), which were separated by a long pause (640 ms; Fig 4A). Live performance of the whistle sounds was often presented to the birds (see below) to facilitate imitation. To human ears, the live sounds are nearly identical to the sounds in the recording. More importantly, the two sources of the melody seem to be identical to the birds, because they sang songs similarly in response to both the live performance (see S1 Movie) and the playback sounds (described below).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Example spectrograms showing imitation by the birds.

The model sound (A) and vocalizations by bird PY (B), by bird C (C) and by bird PK (D). As shown in S3 Movie, the birds learned to imitate the melody step-by-step. These recordings were obtained soon after they began to sing the full melody. PK imitated the music, but did not sing in unison as frequently as the other two birds, so this bird did not participate in further experiments of this study.

https://doi.org/10.1371/journal.pone.0256613.g004

Vocal learning of the melody

From early in life, three birds were exposed to the whistle sound sequence up to 30 times per day, either produced live by the experimenter or from a recording of the melody. The birds spent a large portion of each day with humans and were released from their cage for about 1 hour each day to play with humans. This was done to enhance their social relationship with humans, which could facilitate mimicry of the melody by the birds. Therefore, it was not feasible to keep birds inside an experimental box to record their vocal development as has been done in previous songbird studies [67]. Thus, I do not have a complete history for the vocal development of the birds and vocalizations were only recorded when birds produced sounds of interest (e.g. S2 and S3 Movies).

Vocal imitation was established without any food rewards, though the human caregivers often verbally rewarded the birds for their vocal behavior (see S1 Movie). The birds were not trained with any specific methods, such as the model-rival method [9]. However, eventually, the birds were able to imitate the sounds (Fig 4B–4D, S4 Movie). Throughout this process, both live performance of whistles by the human and a recording of the whistle sounds played back from a PC (Fig 5, see below) were used in parallel as the model sounds. A number of songs were recorded in both the live and playback conditions; however, for the analyses in the Results section, only songs recorded with the playback sounds were used. This method ensures that the stimulus sounds were always consistent. Therefore, the active adjustment of vocal timing to a playback sound by the birds was quantitatively examined.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. A schematic diagram of the updated recording system.

All recordings of singing by the birds used for the analyses in Results section were obtained using this system. Therefore, the stimulus sounds presented to the birds were played back from the PC, not live performance by the experimenter.

https://doi.org/10.1371/journal.pone.0256613.g005

Initial recordings via a single channel

A number of solo songs from the birds and vocal responses to the playback melody were recorded via the monaural channel of the recorder. In the latter phase of development (after the birds were nearly one year old), the recording began to include a considerable number of songs sung in unison with the melody. However, eventually it was difficult to sufficiently separate the vocal signals from the playback sounds in the single channel recordings for analysis. Nevertheless, human listeners could clearly identify the birds singing in unison (S5 Movie, for example). Therefore, a new recording system was built, which allowed for more rigorous analysis.

Recordings with the updated system

Vocal sounds from the birds were recorded by the left channel of the PCM recorder via the microphone, which was directed toward the subject. The stimulus sound signals were sent out from Windows Media Player on a PC. The output was divided into two streams via audio connectors. One of those streams was amplified and presented to each bird as the sound stimulus through the loudspeaker (about 60 dB at the bird’s position) located at the insensitive direction of the microphone; the other was connected directly to the right channel of the PCM recorder. This setup enabled recording of both the birds’ vocalizations and the playback stimulus simultaneously and separately (Fig 5). All data shown in the Results section was recorded with this updated system. There were no other birds in the recording room, which was kept in silence (the background noise level was about 35 dB) during the experiments. Data used for analysis was recorded from each bird at the approximate ages of 330 dph (bird C) and 450 dph (bird PY). Recordings took place 1−2 hours a day for 6 days in total for each bird.

Analyses

Birds’ vocalizations were edited and analyzed with SASLab. Low frequency sounds were cut off with a high-pass filter (0.5 kHz). The sound amplitude was normalized to 75% of the dynamic range at the peak amplitude of the sound sequence. The sound spectrograms (FFT length: 512, Temporal resolution: 87.5% overlap) were created and the onset of each note was obtained with a function of the software (Automatic parameter measurement).

Statistics

Statistical analyses were performed using R3.4.1. Functions “cor.test” with Pearson’s correlation coefficient, “fisher.test” and “wilcox.exact” were used. All tests were performed two-sided.