14
$\begingroup$

Like humans, many animals have unique voices through which we can identify them individually. This information is beneficial in non-invasive tracking, population/density estimation and much more. But the identification process is not very robust in most cases. Therefore to what extent it can be identified is a critical context. Like, known wolf howls can be segregated with 100% accuracy, and wild wolves can be identified with up to 75% accuracy. But the methodology is still in the earliest stage but needs revision. enter image description here

Similarly Gibbon, Hyena have unique voices. What are the other species that have a unique voice?

$\endgroup$
2
  • 3
    $\begingroup$ Where does the picture come from? Would you mind including a reference? (I'm assuming here that you didn't create it just for this question.) $\endgroup$
    – J W
    Jul 27, 2022 at 13:56
  • 1
    $\begingroup$ I suggest this recent review answers the question well for mammals: Linhart et al (2022) "The potential for acoustic individual identification in mammals" link.springer.com/article/10.1007/s42991-021-00222-2 (I am a co-author - but the credit for the work goes mostly to the leading authors) $\endgroup$
    – Dan Stowell
    Aug 4, 2022 at 16:41

5 Answers 5

8
$\begingroup$

What animals?

I'll start with a broad unsupported hunch: probably almost any animal? The details and success are however likely to be a matter of which features (not easy to know beforehand in an understudied species) you use and the stats/ML methodology. When the OP talks about identification, I'm hereon assuming they mean automated identification.

For instance this review by Carlson, Kelly & Couzin 2020 points out various animals that show individual call production (based on evidence that others can identify them only through sound). In their paper you see the literature has evidence of individual recognition in birds, and much smaller mammals (like marmots!). I know there's some evidence for bats having unique echolocation calls too in Yovel et al. 2009 (I say some evidence because N=5 bats if memory serves right).

Hunch: In cases of failure to detect individual signatures, it may be genuine lack of variation or IMHO our genuine lack of understanding on the features to use (esp. for short sounds like clicks or sweeps of a few micros to milli-seconds duration). Here too we might expect to see a vertebrate/mammalian bias. A lot of the voice recognition work is inspired by and based on work in human biometrics.

To what extent?

The question about the stats/ML methods: are the studies using a 'closed-pool' (the 'easier' case) or an 'open pool' (the 'tougher' case) to test the effectiveness of recognizability? A closed pool means there were 10 individuals that form the sample and recognition is always checked by presenting the algorithm with one of the 10 already 'seen' individuals. An open-pool is when 10 individuals were used for training, but the presented individuals can also be 1 previously 'unseen' individual. Here I remember reading that methods used for human voice recognition (based on Gaussian Mixed Models and others) typically show a quality-quantity tradeoff (Encyc. Of Cryptography and Security) . The methods can tell individuals apart in small groups easily but struggle with larger groups. Though a vocalisation may seem stable, it is important to consider ontological change (as animals age the voice change) as well as short term changes (blocked nose, etc).

My suspicion is that many animal studies often have (understandably, just given the insane logistics of training and working w animals!) low sample sizes (few individuals tested), allowing stats/ML techniques to show 'good performance'.

The cues that the stats/ML tools pick up may not always correspond with what the animals are using. Recording animals consistently and 'equally' even in lab settings isn't easy (let's not talk about field recordings now). For instance, a consistently 'loud' animal can be easily identified by eye or algorithm, but is this because of a unique favourite perch position close to the mic? If the identification is for census purposes this is a moot point perhaps, but it may have unintended consequences:). The classifier may sometimes tune in on other consistent cues like stream noise, other adjacent animal calls instead of the actual target species' call! (see Stowell et al. 2018 for a demo on pipelines and things to be wary of while training a classifier).

Note: the OP's question was admittedly asking for other species aside from the one mentioned in the question. The answer here intentionally deviates from creating a list as it doesn't add too much in terms of general know-how on this SE. Also refer to this Meta discussion and participate!          

$\endgroup$
0
5
$\begingroup$

I see two aspects

  • animals emit signature call (as do bottlenose dolphins) then you can easily identify them by the call. This may also be the case for the wolf playback experiment in the OP.
  • animals generate sound with a apparatus that can show individual anatomic differences so that humans can recognize a difference in the calls, then one should expect that sophisticated AI tools can and will recognize individuals (classify to individual level).

In how far one would be able to recognize the cicada or other another insect in your backyard from the others, others may answer, for me they sound all the same, so I do not expect that recognition of individuals is easy or even possible.

$\endgroup$
3
$\begingroup$

Individually specific signatures were described by many researchers in mammals, birds, amphibians, fish... Me and my colleagues, we provided an overview vocal individual identity signatures found across mammals: Linhart, P., Mahamoud-Issa, M., Stowell, D., & Blumstein, D. T. (2022). The potential for acoustic individual identification in mammals. Mammalian Biology. https://doi.org/10.1007/s42991-021-00222-2:

In mammals, HS greater 7 has been found in 23 out of 130 (c.a. 18%) species with the species belonging to Rodentia (9), Primates (8), Carnivora (3), Chiroptera (2), and Cetacea (1) orders.

HS stands for Beecher's information statistic taken as a metric of individuality in vocalizations (see the paper for details). Unlike classification success, this metric is robust to number of individuals in the study and thus is a better metric to compare how easy it is to tell individuals apart in different species.

Besides signature strength, it is also important for some applications that signatures stay consisten over time which is investigated much less often:

Our review indicates that Primates and Cetacea are likely good candidates for successful application of AIID, because they involve species with both strong and stable acoustic signatures.

However, information about stability of individual signatures is largely missing in Carnivora and Chiroptera, taxonomic groups which also possess strong identity signatures. Also, more studies should focus on rodents to > find out if and under which conditions their individual acoustic signatures remain > stable over time. Such studies are crucial for estimating feasibility of AIID and > require repeated recordings of known individuals.

Best wishes, Pavel

$\endgroup$
2
  • $\begingroup$ While I do not disagree with the possibility to recognize individuals by acoustic means, the cited publication gives IMHO only a statement that this should be possible for more than the few species where an association individual vocalisation was possible. Interesting is that for bottlenose dolphins that emit signature whistles the cited publication quotes "Individuality could not be extracted directly from the AIID study on bottlenose dolphins" despite highest HS value. So, saying "Cetacea are good candidates for successful application of AIID" is more a projection than reality. $\endgroup$
    – WMXZ
    Aug 8, 2022 at 17:28
  • $\begingroup$ Yes, it the method of acoustic identification was used just in a few studies to collect NEW information about the species (e.g. capture-recapture study). Based on reported individuality values we can estimate other candidate species in which it might work very well. Individuality in bottlenose dolphin signature whistles was not quantified until very recently, probably, because it was so obvious it is massive. Sayigh et al. (2022) did quantify individuality in bottlenose dolphins to find the highest individuality ever reported so far. $\endgroup$ Aug 9, 2022 at 11:43
1
$\begingroup$

Sometimes individual signals are hidden in plain sight. Check out this take on the individual differences between male capercaillie low-frequency calls Hart et al. 2020. Low frequencies better propagate through vegetation and individual calls might be coded in this low frequencies rather than higher parts of the call. Maybe we should check for individual differences in many more species. Greetings from Austria, Robin

$\endgroup$
2
  • $\begingroup$ Welcome @capreolus! Your answer could be improved with additional supporting information. Please edit to add further details, such as citations or documentation, so that others can confirm that your answer is correct. Please avoid providing just links to another resource, and try to provide context in a manner that answers the question. $\endgroup$
    – Thejasvi
    Aug 1, 2022 at 17:25
  • $\begingroup$ Thanks for the advice Thejasvi! $\endgroup$
    – capreolus
    Aug 2, 2022 at 19:27
-2
$\begingroup$

Dr Carol Bedoya is giving a talk on acoustic identification of individual birds, and census. Tomorrow Tuesday Aug 2, 2022.

Now that I am back at my desktop and not limited to my phone and that the source of this info has updated it too, I can add this:

source: [email protected] = Ben Gottesman

on Slack group Interspecies.io

link for more info (added by Ben): https://docs.google.com/forms/d/e/1FAIpQLSc9kKlIrjfa3wX7vbkkv2augjREUQNSFnZ281DkS8sD3hKsbA/viewform

The link is a page on The Cornell Lab of Ornithology.

Am extract from the above link:

The Acoustic Methods Series is your bi-weekly gathering with a vibrant community of people working in passive acoustic monitoring. In these sessions, we invite speakers to present new acoustic methods and analysis tools, and engage in discussions about hot-button topics in our field. This series started several years ago and we’ve now had dozens of fun sessions. It is hosted by the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology, and open to all who are interested in participating. We meet every two weeks over Zoom, usually at 10:30 AM EST. We would eagerly welcome new participants, especially those from different countries in order to foster an open and connected PAM community and benefit from a diversity of experiences and perspectives. If you are interested in checking out an Acoustic Methods Series meeting, please write your name and email below and I will add you to the listserve. You will receive an email with the Zoom link that works for all our meetings.

Our upcoming session is Tuesday, August 2nd, 2022, at 10:30 EST, and will feature a presentation from Dr. Carol Bedoya on acoustic censusing and individual identification of birds.

$\endgroup$
2
  • $\begingroup$ Welcome @sm1! Your answer could be improved with additional supporting information. Please edit to add further details, such as citations or documentation, so that others can confirm that your answer is correct. You can find more information on how to write good answers in the help center. Please avoid providing just links to another resource. $\endgroup$
    – Thejasvi
    Aug 1, 2022 at 17:23
  • $\begingroup$ Hi @sm1, answer scores are anonymous points given by up/down votes by the community. The -ve score may reflect the answer content, as it currently only references another source with no details in itself. Please see the Meta discussion on this matter. $\endgroup$
    – Thejasvi
    Aug 2, 2022 at 5:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.