RECORDING AND IDENTIFYING BAT ECHOLOCATION CALLS
Brian W. Keelan
keelan@warpmail.net
This account describes the process I followed getting set up to record and identify bats. At the time of this writing (Fall, 2013), it required about $1600 of equipment and software (not including a laptop or a tripod) to make high-quality recordings of bat echolocation calls, and analyze the resulting audiospectrograms (sonograms) to yield reliable identifications (within the U.S. and Canada).
Equipment needed to record bat sonograms include the following:
The first piece of hardware needed is an ultrasonic detector. The most recommended unit for the serious student is the Pettersson D240x (ca $1100), available from just one distributor in the U.S., Bat Conservation and Management (BMC). This device contains a time expansion detector, used to generate an analogue sound signal that has been slowed down 10x so it is in the audible range and can be recorded. It also has a heterodyne detector, which allows the signal to be heard in real time. The unit runs on 9-V batteries, which last about 15 hours in the device.
The time expansion audio needs to be converted to a digital signal for analysis and storage. Although some laptops have the necessary audio port and sound card for doing this (though many do not), better quality can be obtaind for a modest additional expenditure. The Behringer UFO202 is an external sound card that accepts the analogue output of the D240X via audio cable, digitizes it, and provides the output conveniently through the laptop's USB port, which also provides power. This unit, which runs about $40, can also be used to digitize music from tape decks and most phonographs.
You'll also need an audio cable to run from the D240X to the UFO202. The audio cable is available at any electronics store. It needs a male 3.5 mm connector at one end and separate right and left male audio connectors at the other end. You may need to cut, or at least not plug in, one of the channels to avoid crosstalk between the heterodyne and time expansion signals (more details are given in the Setup section).
Although the D240X has a speaker through which the heterodyne signal can be heard, headphones should be used for recording, so that the speaker output does not interfere with the time expansion signal. I use very cheap headphones ($5) that are sold for use while exercising; they are small, lightweight, don't block out all ambient sound (desirable). Idealy they should fold flat and should have a long cable or an extension cable (I run a 5-foot extension cable to a splitter, into which two headsets can be plugged; this gives plenty of slack).
If your laptop keyboard is not backlit, you'll need a light to illuminate it. There are many LED lights that plug into a laptop USB port and then are flexible so they can be positioned over the keyboard; a good current example is this unit , which runs $10. I greatly narrowed the angle of illumination with a long piece of black electrical tape, to keep light from spilling onto the laptop screen. If the above link does not work, try searching for "LED light sticks" on Amazon.com.
A last piece of useful gear is a very lightweight tripod on which to mount the D240X, which, unfortunately, does not have a standard threaded port for a tripod screw, so you'll need to attach it with velcro, micro bungie cords, or something similar. The tripod is helpful for getting the detector away from surfaces, like the ground or water, which might reflect the signal to the detector via a slightly longer path than the direct path, producing a delayed signal that can interfere with the primary signal. The detector weighs nothing, so the tripod can and should be extremely small and lightweight, suitable for carrying on a day pack.
The best software for analyzing and identifying bat sonograms is, appropriately enough, Sonobat . Version 2.9.7 is $320 at the time of this writing. There is a much more expensive version available for professionals, which analyzes recorded sequences and automatically identifies the species present, but for amateurs, much of the fun is figuring out the identifications. Sonobat 2.9.7 provides a number of tools to assist you, and with experience, a substantial majority of calls can be assigned with reasonable confidence. More detail follows in the Identification section.
Headphones and the audio cable are plugged into the D240X, which is attached to the tripod. The unit is switched on by rotating the volume knob on the right side. Monitor the headphones to produce very quiet background static (this has no effect on the recording volume, just the listening volume). The settings I use on the D240X are, on the front of the unit, "Normal", "Time Exp", and "High Gain"; and on the back, "Auto", "1.7", "Low Trigger", and "HF".
To check operation at this point, put on the headphones. One channel from the D240X is the heterodyne signal, and the other channel is the time expansion signal. So, in headphones, the heterodyne frequency-shifted sound is heard in one ear in real time, and the delayed time expansion sound is heard later in the other ear (with about 8 seconds delay). Snap your fingers (or very briefly jangle car keys) to create some ultrasound, so that you can experience this effect. Notice that when you do this, the red LED light goes off on the D240X, indicating that it has triggered and is performing the time expansion. If this does not work, carefully check settings on the D240X.
Only the time-expanded signal is recorded, as it contains the full information regarding the signal; the heterodyne signal is just for real-time monitoring (actually, it can also be used as the triggering signal in the D240X for greater specificity, which can be helpful if low-frequency sounds such as insects are causing too much false triggering). So only one input connection should be made to the UFO202, but as there are two wires and two receptacles, there are 4 possible ways of hooking the device up. You'd think it would be easy to predict how it should be connected, but we've not had much luck with this, and with only four possibilities, it may be faster just to try each arrangement in turn and see which works (see below). So pick one at random and proceed. The small switch near the top left corner should be set to "Line", i.e., pushed towards the corner. (The phono setting is used for signals requiring pre-amplification, such as when digitizing records from a phonograph.)
Next, insert the USB connector into a USB port on the laptop, and keep your fingers crossed that it is recognized as an input to your computer. You may see a pop-up or a dialogue box that will help with doing this; however, you may have to go to your Control Panel to set it as the default audio input for your laptop, and to set the input gain (volume) to record properly.
The process of setting the gain/volume/level is discussed in greater detail in the SonoBat documentation, but in brief, you will need to reach the Microphone/Recording settings screen in the Control Panel. Start the Sonobat AutoRecorder utility, click the button "Monitor without Saving", and look at the trace. As long as the D240X has not triggered (red LED still on) due to some ambient ultasound, the trace should be flat, and the Sonobat "Sound Input Level" gauge near the bottom of the screen should be registering a very low signal (likely under 10 counts). Now snap your fingers, and see if some small amount of background noise becomes evident (typically tens to hundreds of counts). Then watch for the main signal about 8 seconds later, which sould be thousands of counts. If you do not get this behavior, try a different arrangement of audio connections to the UFO202. Once you get the right audio wire into the right connector on the UFO202, adjust the gain so that the peak signal on the SonoBat guage, which you have to watch intently, is over 15,000 counts.
Write this gain down in case your system does not remember it when the laptop is restarted. With the final gain setting, measure the counts when the D240X is not sending (red LED on), and then snap your fingers and measure the counts of background noise when the D240X is sending (LED off, but before the main sound comes through). Set the autotrigger threshold by dragging the green arrow on the guage. It should be set about halfway in distance between the two counts obtained. For what it is worth, our microphone gain with Windows 7 is set at 5 on a scale of 100, and our threshold in the autorecorder is set at 20. You should now be ready to record.
While it is perhaps presumptuous for me to instruct others on how to identify bat sonograms, given my very limited experience, I will still provide an outline of the several approaches I take.
I use a spreadsheet I wrote that, based on tables provided with Sonobat, and output parameter files from Sonobat, computes how many standard deviations from the mean each of 9 characteristic parameters are, for each of the species in California. The squares of the standard deviations are summed for each species to give a single overall metric, and these values are quickly scanned to identify which species are the top contenders. Usually this quickly limits the possibilities to a few species. Furthermore, the standard deviations for each parameter can be checked to determine which characteristics might be most helpful in a visual comparison with reference calls.
Visual Comparison with Reference Calls
Sonobat makes it easy to compare to files of representative reference calls, to see which species might approximately match a call in question. I renamed the composite reference files (starting with two underscores) to begin with their characteristic frequency, so that they automatically sort from low to high frequency, making it even easier to scan through the best options quickly. I also generally consult the extended compiled sets (beginning with a single underscore), and often peruse the individual recordings, because that is the only way to see the full range of behavior for each species.
The more time I have spent studying individual reference calls and their variation, the more I have found it necessary to summarize the information gleaned in a compact way. As a serious amateur botanist, I naturally turn to dichotomous keys to solve this problem. Below I have included my constantly evolving key that I use for distinguishing the central and northern California bats. I recommend that you make similar keys for bats that might occur in your area.
Abbreviations:
Fc = characteristic frequency in kHz
Fm = frequency of maximum power in kHz [Sonobat: f@max]
Fl = lowest frequency in kHz [Sonobat: lo f]
Fh = highest frequency in kHz [Sonobat: hi f]
Su = upper slope in kHz/ms [Sonobat: HiF-Kn]
Sl = lower slope in kHz/ms [Sonobat: Kn-Fc]
Key to Bat Echolocation Calls of Central and Northern California
Main Key
1. Fc > 34.5 ... Key C
1' Fc < 34.5
2. Fc < 15.5 (very rarely MYTH can be as low as Fc = 13; see 2')
3. Social calls, sloped linear to curved, usually with second harmonic, with echolocation calls ... ANPA
3. Echolocation calls, sloped linear to flat, often lacking harmonics
4. Duration < 7.2 ms ... EUMA
4' Duration > 7.2 ms ... EUPE
2' Fc > 15.5
5. Calls concave downwards and/or with moderate ascending sections in first 2/3 of call ... COTO
5. Calls lacking ascending portions, or just with abrupt rise at start, or slight ascent at end of call
6. Calls flat, or descending (usually with curvature) to a flat or low-slope region at end (can be short)
7. Calls essentially flat ... Key A
7' Calls distinctly descending ... Key B
6' Calls steeply sloped throughout or nearly so
8. Calls essentially linear
9. Often having strong second harmonic, power peak near center, Fc = 18-32 ... COTO
9' Lacking harmonics, power peaking near end, Fc = 24-34 ... ANPA
8' Calls with at least a little curvature, and/or a hint of a tail, and/or Fh > 76
10. Calls steeply sloped right to end, with Fh-Fl > 40.5 and/or drooping/downcurved tail
11. Fc > 29.5 (rarely down to 29) ... MYEV
11' Fc < 29.5 (rarely up to 32) ... MYTH
10' Calls becoming less sloped towards end, with Fh-Fl < 40.5 and tail absent ... ANPA
Key A: Flat calls, Fc = 15.5-30
1. Having very sharp leading upsweep and/or trailing downsweep (can be subtle) ... TABR
1' Having leading downsweep (rarely indistinct upsweep) and/or trailing upsweep
2. Fc > 22.5 ... LANO
2' Fc < 22.5 ... LACI
Key B: Calls descending (usually with curvature) to a flat or low-slope region (can be short), Fc = 15.5-34.5
1. Having trailing downsweep w/ Fc-Fl > 3.5 kHz, and/or leading upsweep (can be almost vertical) or plateau
(may be short -- check in standard mode); never upswept at end; features indistinct in weak calls ... TABR
1' Normally descending right from start, though can become flat or minutely drooping at the end of the call
2. Fc rarely above 26, concave throughout, sometimes with sharp leading downsweep, but no knee ... LACI
2. Fc rarely below 25, end can be essentially flat, lacks sharp leading downsweep, can have a knee
3. Power profile peaking late or multi-lobed; minimal flat area at end; can be linear above
distinct knee; can have <6 calls/second ... ANPA
3' Power profile unlobed, fairly symmetric; up to 3/4 of call nearly flat; curved above knee if
one is present
4. Fc > 29.5 and/or Fh > 54.5 and/or end of call slightly upswept ... EPFU
4' Fc < 29.5 and Fh < 54.5
5. Fh < 37 and/or end of call sometimes minutely drooping (Fc-Fl = 2-3 kHz) ... LANO
5' End of call shallowly concave upward to linear ... LANO/EPFU (ca. 2/3 of calls will key here)
Key C: Fc > 34.5
1. Call with gentle power profile, lacking any tail, sigmoidal or with asymptotically flat region
2. Fc rarely > 45, ms rarely < 5; call smooth, sometimes sigmoidal; power profile fairly uniform; Fc can shift between calls ... LABL
2' Fc never < 41, ms rarely > 9; call often with sharp kink before power peak, after which call essentially flat ... PIHE
1' Call with sharp, late power peak (except some MYLU), followed by an inflection point, usually with a drooping tail; no asymptotically flat region
3. Call starting with distinct flat to ascending section ... MYVO
3' Call initially descending
4. Call with Fc > 45 usually, often > 50, never < 43
5. Power building smoothly to late peak, then dropping quickly; often with sharp knee; tail short and straight; usually over water ... MYYU
5' Power profile multi-lobed; lacking sharp knee; tail long and drooping; near canopy; contraindications: ms > 6, Su < 16, Sl < 3 ... MYCA
4' Call with Fc < 45 usually, often < 43, never > 50 (4 difficultly separated species: MYME, MYEV, MYLU, MYVO)
6. Calls smoothly concave with sharp inflection at Fc, angling into a long (Fc-Fl > 4), straight, tail; bandwidth high
7. Fc > 39; often near peak power for well over 1 ms ... MYME
7' Fc < 39 (Fc can be higher, but then calls nearly linear) ... MYEV
6' Calls without such distinct tails, or more linear, or less smooth (with bends or kinks), or lower in bandwidth
8. Calls with high bandwidth, steep, and nearly linear
9. Fc <= 43, can be < 34 ... MYEV
9' Fc >= 34, can be > 43 ... MYVO
8' Calls with low bandwidth and/or substantialy curved
10. Calls long with low slope: Su < 5; or St < 4; or ms > 6 with Sl < 3; power profile often multi-lobed ... MYLU
10' Calls generally shorter and/or more sloped; ofetn with simple power profiles ... MYVO/MYLU/MYME/MYEV (many calls will key here, unresolved)