Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Brief Communication
  • Published:

Reconstruction of vocal interactions in a group of small songbirds

Abstract

The main obstacle for investigating vocal interactions in vertebrates is the difficulty of discriminating individual vocalizations of rapidly moving, sometimes simultaneously vocalizing individuals. We developed a method of recording and analyzing individual vocalizations in free-ranging animals using ultraminiature back-attached sound and acceleration recorders. Our method allows the separation of zebra finch vocalizations irrespective of background noise and the number of vocalizing animals nearby.

This is a preview of subscription content, access via your institution

Access options

Rent or buy this article

Prices vary by article type

from$1.95

to$39.95

Prices may be subject to local taxes which are calculated during checkout

Figure 1: A wearable sound and acceleration logger for recording individual vocalizations in a group of songbirds.
Figure 2: Acceleration allows for reliable segmentation of vocalizations.
Figure 3: Vocal interactions in a group of songbirds.

Similar content being viewed by others

References

  1. Yartsev, M.M., Witter, M.P. & Ulanovsky, N. Nature 479, 103–107 (2011).

    Article  CAS  Google Scholar 

  2. Barchi, J.R., Knowles, J.M. & Simmons, J.A. J. Exp. Biol. 216, 1053–1063 (2013).

    Article  Google Scholar 

  3. Sober, S.J. & Brainard, M.S. Nat. Neurosci. 12, 927–931 (2009).

    Article  CAS  Google Scholar 

  4. Andalman, A.S. & Fee, M.S. Proc. Natl. Acad. Sci. USA 106, 12518–12523 (2009).

    Article  CAS  Google Scholar 

  5. Grohrock, P., Häusler, U. & Jürgens, U. J. Neurosci. Methods 76, 7–13 (1997).

    Article  CAS  Google Scholar 

  6. Hiryu, S., Hagino, T., Riquimaroux, H. & Watanabe, Y. J. Acoust. Soc. Am. 121, 1749–1757 (2007).

    Article  Google Scholar 

  7. Hiryu, S., Shiori, Y., Hosokawa, T., Riquimaroux, H. & Watanabe, Y. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol. 194, 841–851 (2008).

    Article  Google Scholar 

  8. Vyssotski, A.L. et al. J. Neurophysiol. 95, 1263–1273 (2006).

    Article  Google Scholar 

  9. Vyssotski, A.L. et al. Curr. Biol. 19, 1159–1166 (2009).

    Article  CAS  Google Scholar 

  10. Elie, J.E., Soula, H.A., Mathevon, N. & Vignal, C. J. Acoust. Soc. Am. 129, 4037–4046 (2011).

    Article  Google Scholar 

Download references

Acknowledgements

The study was supported by research grants from the Scientific & Technological Cooperation Programme Switzerland-Russia (A.L.V., A.V.L. and V.N.A.), the University of Zurich (A.L.V.) and the Swiss National Science Foundation 31003A_127024 (R.H.R.H.), and by the European Research Council under the European Research Community's Seventh Framework Programme (ERC grant AdG 268911 FP7/2007-2013).

Author information

Authors and Affiliations

Authors

Contributions

V.N.A. and A.L.V. performed the experiments and analyzed data. J.A.H. wrote software for real-time song detection and synchronization of the loggers. A.N.A. wrote logger firmware. A.V.L. and R.H.R.H. supervised the project and wrote software for the data analysis. A.L.V. designed the study and wrote the paper together with R.H.R.H.

Corresponding author

Correspondence to Alexei L Vyssotski.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Integrated supplementary information

Supplementary Figure 1 Adaptation of Neurologger 2A for sound and acceleration recordings

(a) Top view of the neurologger printed circuit board (PCB) with microphone (MIC) and accelerometer (ACCEL) attached. (b) Bottom view of Neurologger 2A PCB. Components that need replacement are marked in red. Wires (red lines) connect microphone and accelerometer with the board. Wire connects (soldering points) are marked by red dots. Input pads Ch0 and Ch1 are connected with the microphone signal output, and Ch2 and Ch3 with the accelerometer signal output; +1V is the power supply for the microphone and the accelerometer. Microphone and accelerometer are glued to the board by two component epoxy adhesive (DOUBLE/BUBBLE® Epoxy, www.hardmanadhesives.com).

Supplementary Figure 2 Attachment of the harness to a zebra finch

The bird is anesthetized with Isoflurane (3% induction, 1.5% support) delivered through a silicon tube going to the beak. The harness (Fig. 1a) was fabricated in advance from a piece of Velcro strip of 15x22 mm and a BüroLine rubber band of 40x1.3 mm (www.bueroline.com, Cat. #155012). The rubber band is sewn to the Velcro strip at two places, the knots of the thread are fixed by cyanoacrylate glue. (a) The harness is placed on the bird; the loops of the rubber band on the chest are attached to each other with a piece of thin laboratory/surgical rubber glove. (b) The rubber endings are glued together with a drop of cyanoacrylate, a small piece of paper is placed underneath to prevent gluing the feathers. (c) Bottom view of the attached harness. (d) Top view of the attached harness. The tightness of the harness is adjusted via the length of the rubber ring on the chest. It was found that tension of the harness has little impact on the quality of sound recordings. The tension should be such as to prevent any unnecessary pressure on the animal and to prevent sliding of the backpack on the bird’s back; a loose harness can disturb the animal as much as an overly tight harness. To attach the logger onto the Velcro we recommend placing forceps between the animal’s back and the harness to push against the forceps and not against the back. To remove the logger we recommend holding the Velcro strip of the harness in one hand and the backpack in the other hand.

Supplementary Figure 3 Spectrograms of a song motif recorded with the wall microphone and the accelerometer (bird g2k8)

(a) Spectrogram of a song motif recorded with the wall microphone. (b) Spectrogram of the same song motif recorded with the accelerometer. Abbreviations: “T” - tet calls, “I” - introductory notes, “A”, “B” and “C” – song syllables. The low-frequency bumps in acceleration that tend to occur before syllable onsets (black arrows) are presumably associated with rapid changes in air sac pressure (respiration-related thoracic movement).

Supplementary Figure 4 Habituation of old, young and juvenile animals to the chamber and the backpack

Old (n = 6) and young (n = 3) animals were habituated to the sound proof individual isolation chambers until stationary amounts of movement and singing were observed. Then, the backpacks were attached (day zero), backpacks were attached to old animals after 5 days and in young animals after 9 days. (a) Shown is the daily amount of movement measured by infrared video camera (motion, expressed in percent of the baseline motion (black dashed line) in old animals achieved on days [-3 -1]). The baseline value (2.27%) is an output of “Motion Detector” (range [0 100], %) averaged over light phases of days. For young animals the baseline motion (red dashed line) was given by the average motion on days [-6 -1] (5.63%). Error bars indicate s.e.m. Asterisks label days when motion significantly deviated from the corresponding baseline level (paired two-sided t-test, P < 0.05). Locomotor activity of young animals exceeded activity of old animals on all days (P < 0.05, non-paired two-sided t-test). (b) Number of song motifs produced per day. Baselines were computed by averaging number of motifs on days [-2 -1] in old birds and on days [-4 -2] in young birds. Asterisks mark significant deviations from baselines as before. Number of motifs produced by old and young animals differed on some days. (c) Number of calls per day. Number of calls in young group increased monotonically during 9 days of habituation to the chamber and did not reach saturation. For this reason we took the value on day -1 as estimate of baseline in young animals. Baseline calling in old animals was estimated as average on days [-2 -1]. Red horizontal bars along time axis indicates significant difference between old and young animals (P < 0.05, non-paired two-sided t-test). Notably, the rate of calling increased in young birds during habituation to the backpacks (days [1 9]) in roughly similar manner as during habituation to the chambers (days [-9 -1]). This means that the severity of calls disturbance in young birds caused by the backpacks is similar to the disturbance caused by the isolation chambers. (d) On two juveniles we attached 2.0 g backpacks on post-hatch-days 39 and 46 (time zero on x-axis). In the first juvenile (black) the backpack hardly decreased singing rate; in the second juvenile (red) the singing rate was transiently suppressed and picked up two days later. Both birds consistently produced more than 10’000 song syllables and introductory notes per day just 3 days after backpack attachment, suggesting that our recording technique is permissive of song learning studies in juveniles.

Supplementary Figure 5 Non-vocal response to song

(a) An example of non-vocal response in co-singing birds 3 and 4. The song oscillogram (blue) in bird 4 was recorded with the wearable microphone. The non-vocal response (body acceleration, red) in bird 3 is reflected in the root-mean-square (RMS) accelerometer signal (sliding window) in the non-vocal low-frequency band < 300 Hz. Bird 3 did not vocalize during this episode (not shown). (b) Normalized median RMS traces of body acceleration in all four birds aligned to the beginning of the song motif in bird 4 (onset of syllable A). Shown are the median filtered RMS traces of 72 motifs. We selected only episodes in which no vocalizations in birds 1 - 3 were detected within ± 2 s from motif onset in bird 4. The 3 SDs level (red dashed line) in bird 3 was estimated using bootstrapping. Note that significant (> 3 SDs) movements in bird 3 took place in a narrow interval -217 to 40 ms around the onset of the motif. Possibly, bird 3 reacted to the preceding introductory notes or to syllable C in the previous motif. By contrast, bird 4 did not show any detectable non-vocal response to the songs of bird 3, indicating asymmetry in movement responses in these two animals.

Supplementary Figure 6 Accelerometers achieve high–signal-to-noise recordings of vocalizations irrespective of background noise level

(a-g) SNR Sound pressure levels (SPL) achieved with the wall microphone (light gray bars), backpack microphone (medium gray bars), and acceleration SNR (dark gray bars) during production of different calls, introductory notes, and song syllables in bird g2k8. The four levels of white background noise were: 0, 50, 60 and 70 dB. SNRs vary in the range [-23.1 37.9] dB for the wall microphone, in the range [0.92 33.9] dB for the backpack microphone, and in the range [11.6 29.3] dB for the accelerometer, showing that accelerometers provide the only feasible approach to unconstrained recordings of vocalizations under noisy conditions. SPLs are shown relative to 2 × 10-5 Pa (international standard). Acceleration is shown relative to 10-3 m/s2. (h) Intensity of background noise recorded with microphones and accelerometers. SNRs in a-g were computed relative to these levels.

Supplementary Figure 7 Wearable accelerometers allow discrimination of vocalizations from four males

The upper panel shows a sound spectrogram of vocalization of four males to be separated and analyzed (the sound was recorded with a wall-attached microphone, see also the Supplementary Video 1 for the birds’ behavior). The spectrograms in the middle show that the wearable microphones pick up mixed vocalizations from several birds, making signal discrimination difficult at least. By contrast, accelerometers spectra (bottom) recorded vocalization of just their hosts, allowing for simple threshold-based vocal discrimination. Episodes of individual vocalizations are marked by yellow horizontal bars. The accelerometers were sensitive to rapid bird movements. During episodes of inactivity accelerometers were often able to pick up heart beats (in animals 1 and 3, red ellipses).

Supplementary Figure 8 Segmentation and classification of all vocalizations in a bird group

(a) Shown are spectrograms of example vocalizations in individual birds recorded with back-pack microphones. Song motifs consist of syllables labeled “A”, “B”, and “C”, and introductory notes labeled “I”. Only bird 1 produced distance calls (such calls are relatively rare in domesticated populations of zebra finches). (b) A 2.5-h temporal sequence of morning vocalizations. The zero time point marks the lights-on event at 7 a.m. Each vocalization (distance-call, call, introductory note, syllables: A, B, C) is represented by a vertical line of corresponding color. (c) A magnified 25-s episode taken from (b) during which all four animals sung. This interval corresponds to the spectrogram shown in Supplementary Figure 7.

Supplementary Figure 9 Vocalization spectra measured with backpack microphones and accelerometers

(a-d) show median vocalization spectra (solid lines) of birds 1-4 recorded with microphones #1 to #4. Spectra were computed from 166-ms fragments starting with vocalization onsets (calls, introductory notes and syllables). Only vocalizations that did not overlapped with vocalizations or noises (e.g. wing flaps) produced by other animals were selected. In birds 1-4 the numbers of such vocalizations were 972, 4535, 3344, and 1468, respectively. Dotted lines represent background noise recorded by microphones when no bird vocalized. To compute spectra of background noise, 10000 166-ms sound fragments during quiescence were randomly taken. Note that sound intensity at the microphone of the vocalizing bird exceeded background noise by about 30 dB. Sound intensity at microphones of listeners exceeded background noise by about 20 dB. Spectra at the microphones of listeners were attenuated but similar to spectra at carrier microphones. SPL values are given relatively to 10-6 Pa/. (e-h) show median vocalization spectra (solid lines) of birds 1-4 recorded with accelerometers #1 to #4. Dotted lines represent spectra of background noise when no bird vocalized. The peak in the vicinity of 8 kHz is caused most probably by resonance of the accelerometer sensor (luckily, the resonance frequency lies above the frequency range of interest). Note that the spectra of the listeners essentially does not differ from their background noise spectrum. The spectra of the vocalizing birds exceed the background noise spectra by about 20 dB at 1 kHz. Acceleration values are provided relative to 10-5 m/s2/.

Supplementary Figure 10 Vocalization spectra of bird 3 when it is close to bird 4

(a) Vocalization spectrum recorded with microphones. Note that microphones practically do not discriminate vocalizations in proximal birds 3 and 4 (red and blue lines almost coincide). Dotted lines denote background noise. Shown are spectra of 274 vocalizations in bird 3 during which the estimated distance between the two birds was ≤ 5 cm (the distance estimate was based on relative sound amplitudes between singer and listener ≤ 3 dB). (b) Vocalization spectra of the same vocalizations recorded with accelerometers. Note that the spectrum on the vocalizer accelerometer exceeding that of all listeners by about 20 dB (at 1 kHz).

Supplementary Figure 11 Cross-correlation (CC) and autocorrelations of songs

Correlations were computed after summing the number of song vocalization onsets in 50-ms sliding windows and subsequent convolution with a Gaussian kernel of width σ = 20 ms. (a) The CC function of song vocalizations of co-singing birds 3-4 has a wide peak in excess of 3 STDs. Shown is the same curve as in Figure 3e but without smoothing. (b, c) Autocorrelations of calls in birds 1 and 2 (b) and 3 and 4 (c). The autocorrelations exceeded 3 standard deviations (SDs, dotted lines of corresponding colors) in the relatively wide interval [-3 3] s. In some birds, high autocorrelation peaks were produced by complete motif matching and lower peaks by matching of non-identical syllables (b). The lower CC values in a compared to autocorrelations in c reveal that co-singing happened less frequently than production of two consecutive motifs by a single bird.

Supplementary Figure 12 Temporal alignment of backpack records

(a) Drifts of internal clocks on three loggers are shown by blue, green and magenta lines relatively to the fourth logger taken as a reference (turquoise horizontal line). Drift was measured by comparing patterns of IR pulses stored on loggers. A Labview program monitored sound recorded with a wall-attached microphone, every 4 ms it assessed whether a song was present, in which case a 0.8-ms IR pulse was sent to the loggers for synchronization. Temporal drift was always linear except in one logger (shown in green) in which a couple of sudden interruptions were observed for a dozen of milliseconds (black arrows). Logger records were aligned by zeroing the lines: we stretched or compressed the records by adding (linear interpolation) or removing data points. (b) Shown are residuals after linear regression of clock drift curves in a (because of the two interruptions we performed linear regression on 3 segments in logger 3). Most of the time residuals did not exceed 1 ms. Only during the first 0.5 h residuals were relatively large (< 3 ms) because of low density of IR pulses caused by infrequent singing. More frequent emission of randomized IR pulse sequences could eliminate this problem. The root-mean-square deviation was 0.46 ms for the whole record and 0.25 ms after omission of the first 0.5 h. (c) Drifts of internal clocks like in a but measured by temporal matching of sound envelopes recorded with the logger microphones. See Online Methods for details. (d) Residuals of sound-based synchronization similar to b. The root-mean-square deviation was 0.37 ms for the entire record and 0.35 ms after omission of the first 0.5 h. Thus, synchronization based on sound works as well as synchronization based on IR pulses and may be useful outdoors where IR synchronization can be difficult.

Supplementary Figure 13 Correlations between calls, introductory notes and identified syllables

(a) Correlation diagram in the pair 3-4 that tended to co-sing. The color encodes the PCCs and white asterisks indicate their significance (*P < 0.05, **P<0.001). Note that all song vocalizations (syllables and introductory notes) are positively correlated to an approximately equal extent. (b) Correlation diagram between different vocalizations (calls, introductory notes and syllables) in pair 2-4 that avoided singing together. Red arrows and red asterisks indicate significant difference2 between PCCs. Animal 2 is able to discriminate syllables A-B and A-C in the song of animal 4: it begins to sing (an introductory note) much less frequently during syllable C than during syllable A.

Supplementary Figure 14 Sound amplitudes of vocalizations measured with near and far backpack microphones and distances between animals during songs and calls

Four panels (a-d) show median sound amplitudes during vocalizations of birds 1-4 recorded with microphones 1-4. Standard errors are estimated by bootstrapping. Sound amplitudes are reported in units of air pressure mPa (1 mPa = 10-3 H/m2). Note that on average song vocalizations were 3.16 ± 0.80 times louder than calls (amplitude, mean ± s.e.m). The ratio of amplitudes on vocalizing and listening birds’ microphones was used to estimate the distance between animals, assuming a fixed distance between the beak and the backpack microphone in the vocalizing bird (3 cm). The four panels (e-h) show median distances between vocalizing birds 1 - 4 and all listeners when the vocalizing animal was either calling or singing. The shown standard errors were estimated by bootstrapping. Note that bird 1 kept a large and similar distance with all other birds during calling and singing (e). Bird 2 was closer to other birds during calls, but when singing it was closer to bird 1 and further away from bird 4 (f). The co-singing birds 3 (g) and 4 (h) were closer to other animals during singing than during calling. *P < 0.05, **P < 0.001, bootstrap.

Supplementary Figure 15 Self consistency of sound amplitude ratios validates distance estimation approach

Six panels (a – f) represent sound amplitude ratios on calibrated microphones in all bird pairs: 1-2, 1-3, 1-4, 2-3, 2-4, and 3-4. Each dot represents two notes (produced by two different birds) separated by a small time interval (<0.6 s between note onsets). During these quasi-synchronous notes we considered the positions of animals as fixed. The horizontal axis reports the square root of the ratio S*i/Sj of microphone amplitudes, where the amplitude S*i on the vocalizing animal’s back-attached microphone is marked by an asterisk. The vertical axis displays the corresponding ratio when the other animal is vocalizing. The coefficient bij is the regression coefficient estimated by minimizing the sum of squares of residuals ɛk, k = 1,…,N (see panel a), where N indicates the number of quasi-synchronous notes. Coefficients bij are close to 1, indicating validity of our calibration approach, but significant spread of points reveals low accuracy of one-shot measurements. For distance estimation we assumed that both backpacks had equal distances to the beaks of carrying birds and that these distances were much smaller than the distances between animals.

Supplementary information

Supplementary Text and Figures

Supplementary Figures 1–15, Supplementary Tables 1–3 and Supplementary Note (PDF 2566 kb)

Four male zebra finches singing together

The movie demonstrates complexity of vocalization separation in a group of birds. The proposed back-attached sound and acceleration recorders are able to record individual vocalizations even in such challenging environment. (AVI 4422 kb)

Juvenile zebra finch singing with the backpack

The movie shows that the weight of backpack is an acceptable burden even for a 40 days-old juvenile bird. Ability to sing with the backpack permits investigation of song learning in social settings. (AVI 2463 kb)

Juvenile zebra finch flying with the backpack

This movie shown that the backpack does not drastically change the locomotor activity of juvenile bird. This suggests that this load will not affect natural behaviours of the bird. (AVI 2432 kb)

Behaviour of juvenile zebra finch before backpack attachment

This movie demonstrates behavior of 39 days-old bird. It is given for comparison with the Supplementary Videos 2 and 3 that show behavior of the same bird with the backpack one day later. One can note that the behavior is almost unchanged when the backpack is attached. (AVI 5544 kb)

Vocal interactions in a couple of zebra finches

The movie shows singing male with the female answering to the male song by soft calls. These calls are hardly hearable in the record of the wall microphone, but are easy-detectable by the backpack accelerometer (Fig. 2a). (AVI 2228 kb)

Supplementary Software

The supplementary software demonstrates the data processing done in the article. The code is written in Matlab. In particular, the most computation intensive and unusual part of data processing is shown – synchronization of data records (i.e. acceleration and sound signals on backpack, as well as sound on wall microphone). Sample dataset is provided for software testing. The detailed description of software is provided in Description.pdf included in the archive. (ZIP 26737 kb)

Source data

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Anisimov, V., Herbst, J., Abramchuk, A. et al. Reconstruction of vocal interactions in a group of small songbirds. Nat Methods 11, 1135–1137 (2014). https://doi.org/10.1038/nmeth.3114

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/nmeth.3114

This article is cited by

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing