Dolphin Chat Project

Dolphin communication captivated scientists and the public for decades. Even though it has been studied quite extensively and many exciting discoveries have been made (i.e. signature whistles), we still do not fully understand how dolphins communicate and what information their phonations could convey. New exciting research does show that dolphins could indeed coordinate their actions via acoustic communication, so their phonations do convey at least some meaning and we would like to know what is being communicated and how.

The project’s overall goal is to analyze acoustic (and behavioral) data with the help of citizen scientists. Data will be used to train deep learning model that will be afterwards used to help citizen scientists to make faster, more accurate classifications (like we currently do in our other project, Manatee Chat). We are starting with Phase one of this project where citizen scientists are asked to identify whether or not the sound file contains dolphin vocalizations (phonations) and also what other sounds are present in the file so our deep learning models can better understand the auditory scene.

Bottlenose Dolphin Phonations

Bottlenose dolphins produce a variety of acoustic signals that can be roughly divided into two main types: pulsed and tonal. Pulsed phonations include echolocation clicks, clicks trains, terminal buzzes and burst pulses. Tonal signals include whistles, low-frequency narrow band signals (you can read our paper about those), and a number of “other” phonations that have a variety of names in research literature. Bottlenose dolphins have exceptionally wide hearing ranging between 1 and 150 kHz with their best hearing sensitivity between 15-110 kHz. Dolphins also have exceptional temporal resolution allowing them to perceive very rapid signals (something that humans are not able to do). Such exceptional hearing abilities mean that dolphins have very broad and complex vocal repertoire, much of which is actually above human hearing. What this means practically is that you will see some phonations on the spectrogram but will not be able to hear them. We do not know the purposes of many of those phonations or their meanings and this project designed to investigate it.

In our files, you will encounter two broad categories of sound: pulsed and tonal. Pulsed sounds will include echolocation clicks, click trains, terminal buzzes, pops, and burst pulses. Tonal sounds will include whistles, low frequency narrow band sounds, and a variety of additional tonal calls that have been called a variety of names in the literature like squeaks, brays, and others. This is Phase 1 of our Dolphin Chat project, so just like with the Manatee Chat Phase 1, we simply ask you to determine if ANY dolphin phonations are present in the sound file.
Important note: As you will learn in our specialized training, many dolphin phonations are beyond human hearing range, so unlike in Manatee Chat, you more often than not will not be able to hear many dolphin phonations in Dolphin Chat. For that reason, it is very important to examine the spectrogram extra carefully.

Other Sounds

Soniferous Gulf Toadfish (Opsanus beta)(Photo by Steve Ryan)

In order to understand dolphin communication, we should also consider the soundscape these marine mammals inhabit. For that reason, we will ask citizen scientists to identify additional sounds both of biological and anthropogenic (human made) origins. The most common biological noise you will encounter will be from the snapping shrimp, which is a random clicking noise. It could make it challenging to identify dolphin clicks in the presence of snapping shrimp, so please refer to our training, tutorial, and quiz to familiarize yourself with these sounds better. Sometimes you will hear fish sounds in our files, specifically toad fish but also others. Both areas where data was collected have a number of somniferous fish including Cynoscion arenarius, Bairdiella chrysoura, Cynoscion nebulosus, and the toadfish Opsanus beta. Some of these fish sounds are very distinct and easy to identify (e.g. toadfish). Anthropogenic sounds include boat noise and occasional depth/fish sounders that could also be heard in our files. As with the Manatee Chat, we ask you to ignore sounds coming from splashing water and mechanical noises from our recording system, or when the hydrophone rubs against something like the sea bottom, etc.

Recording Format: Narrowband and Broadband Workflows

Broadband file recorded at 192-200 kHz sampling rate

The same file downsampled to lower sampling rate, 48 kHz. Note how different echolocations clicks, burst pulses and whistles look in this file.

You will see two types of sound files in Dolphin Chat (narrowband and broadband workflows). Narrowband workflow will be similar to Manatee Chat, i.e. 48 kHz sampling rate, giving the recording range up to 24 kHz. However, the broadband workflow will have much higher sample rate (192-200 kHz) giving much higher recording range, up to 100 kHz. Sounds will look different in these two workflows and this is important to keep in mind when identifying sound categories. Please refer to our field guide and additional specialized training to learn more.

Data Collection Sites

Data was collected in the Tampa Bay area and also in the Mississippi Sound area. Both habitats have significant differences. Tampa Bay is shallower and has more stable salinity levels (34-35 ppt versus 16-24 ppt in the Mississippi Sound). Tidal cycles also differed, mixed tides in Tampa Bay area and diurnal tides in the Mississippi Sound area. Water depths where dolphins were encountered were on average between 16 to 18 ft and water temperature was between 66F to 71 F. All recordings were made under the NOAA permit #1080 and data collection was during daytime, between 8 am and 4 pm.
Update: We will be adding data from publicly available depositories (Macaulay Library).

Project Significance

Ever since the times of Ancient Greeks, humans have been wondering if dolphins have “language” or at least some sort of sophisticated communication system. This project will bring us a little closer to better understanding what information dolphins’ calls could possibly convey. Additionally, the utilization of deep learning models allows us to process a lot of data, something that has not been possible up until relatively recently. Accurate identification of dolphin sounds will allow better performing passive acoustic investigations as we will be able to identify if dolphins are visiting certain areas based on their acoustic signals alone.

Specialized Training for Citizen Scientists

Following the success of our Manatee Chat practice and quiz and improved accuracy in citizen scientist ratings, we will also be offering specialized training for the Dolphin Chat project as well. Please visit our website and try our practice to identify various sounds and the final quiz where you can test your accuracy and knowledge. We found that by offering training to citizen scientists we could improve accuracy and inter-rater reliability, so we ask citizen scientists working on Dolphin Chat to consider participating in this specialized training.

References

Au, W. W. (2000). Hearing in whales and dolphins: An overview. In Hearing by whales and dolphins (pp. 1-42). Springer, New York, NY.

Au, W. W., & Hastings, M. C. (2008). Principles of marine bioacoustics (pp. 121-174). New York: Springer.

Janik, V. M. (2009). Acoustic communication in delphinids. Advances in the Study of Behavior, 40, 123-157.

Jones, B., Zapetis, M., Samuelson, M. M., & Ridgway, S. (2019). Sounds produced by bottlenose dolphins (Tursiops): a review of the defining characteristics and acoustic criteria of the dolphin vocal repertoire. Bioacoustics, 1-42.

Ridgway, S. H. (1990). The central nervous system of the bottlenose dolphin. The bottlenose dolphin, 69-97.

Simard, P., Lace, N., Gowans, S., Quintana-Rizzo, E., Kuczaj, S. A., Wells, R. S., & Mann, D. A. (2011). Low frequency narrow-band calls in bottlenose dolphins (Tursiops truncatus): signal properties, function, and conservation implications. The Journal of the Acoustical Society of America, 130(5), 3068-3076.