April 2, 2005

HPWREN Team Experiments with Real-Time Data from Automated Wildlife Acoustics Sensors

Originally instigated more than a year ago by the desire to capture wolf howls at the California Wolf Center (http://hpwren.ucsd.edu/news/040107.html and http://hpwren.ucsd.edu/news/050205.html), it became desirable to add an automated acoustics capability to HPWREN sensors. In addition, the work by Stuart Gage of Michigan State University (http://hpwren.ucsd.edu/news/050210.html) at the SDSU Santa Margarita Ecological Reserve has shown that this is a realistic and affordable application, as have discussions with Deborah Estrin and her research group about their work at UCLA (http://cens.ucla.edu).

To evaluate the feasibility of a system for an automated and inexpensive capability to capture sounds from wolves and birds, HPWREN researchers utilized a small embedded XScale-based Linux platform and other off-the-shelf components. The prototype uses a single board Arcom Viper computer. and compared various microphones for the sound input. The platform consumes fairly little electric current and is easily deployable in the field in a solar-powered or Power-over-Ethernet (using a cable splitter) environment.

xscale viper Xscale Arcom/Viper platform used for an acoustics sensor

For an initial use, the system was placed into a cooler box, with electric current and network connectivity provided via a Power-over-Ethernet setting. The microphone was mounted outside of the box, which allowed for some interesting experiments to provide wind shielding and to limit damage caused by rain and wildlife.

acoustics sensor in cooler box deployed acoustics sensor

Above: Deployed acoustics sensor

Left: Acoustics sensor and microphone

The objective was to automate the process of the sound-data collection, distribution to a central server, and to make the results available on a web site, which is similar to the motion-detect images shown at the HPWREN cameras web site.

The audio software in the XScale system uses the "listener" software by Folkert van Heusden, who has been instrumental with the success of this activity, including his modifications to the software so that it is more compatible with this particular application. "Listener" allows certain parameterization, such as sound threshold, sound filter, file format, and file disposition. For the disposition, a shell script is executed for a transfer to a central server and subsequent local file deletion. A program on the central server regularly looks for new files, processes them, and makes them available on a web site. http://grizzly.hpwren.ucsd.edu/Acoustics/Data/MCR/ shows some examples.

Of particular use to this application are audio spectrum analyzers, including Erik Olson's "baudline," and Dave Andruczyk's "eXtace." Both Erik Olson and Dave Andruczyk have been very helpful as well, and provided valuable information.

"Listener," "baudline," and "eXtace" are all freely available software, and run on both Linux and FreeBSD systems.

After configuring the web browser to automatically start the "baudline" audio spectrum analyzer application, a bird song sample captured with "listener" is being analyzed

The images below show graphing capabilities of the "eXtace" audio spectrum analyzer, including in 3-D. The original images were provided by Dave Andruczyk.


Some issues that need to be addressed prior to broader use is microphone survival under adversarial weather and wildlife conditions, and an automated pre-analysis of the collected data. A lower noise floor in the audio hardware and higher microphone sensitivity are also desirable, especially for fairly quiet wildlife monitoring environments.

Some additional analysis using baudline, with the images and quoted comments provided by Erik Olson

Note, the images are clickable for full-resolution versions.

baudline "I zoomed in the color resolution to maximize the gain and extract as much detail as possible. I also adjusted the the Gaussian window beta value to improve the time resolution. Notice the fast pulsing (~12ms) of the the upper chirp. In the lower section notice the high frequency harmonics at around 16kHz. The 16kHz pulse is a 2nd harmonic while the 16 - 20kHz sweep is a 3rd harmonic. Not sure what it means or why it is but I find it interesting. You have a pretty high quality recording setup and these distortions might be useful in ID'ing individual birds. Or the harmonics are caused by the recording equipment. Just guessing."

"The signal looks like it is mono with a slight delay. I did some stereo steering ops and the purple tinge color is directional information. So the purple is likely just the one sample delay. A stereo recording of a moving target would be more interesting. In any case the purple accents the decay of the chirps. I have some future plans of adding a higher quality directional extraction feature, it should be very interesting. Hopefully I can find some funding for it."

baudline "Just a vibrant color palette zoomed way in the time axis."

"The Average, Spectrogram, Spectrum, and Waveform views. You can see the same 16kHz harmonics that were in green.png. The Average window is interesting in that each chirp phrase in the Spectrogram was copy-n-pasted into the Average window as a separate color bank. Just a different way of looking at the chirp sections. Might be useful for ID."

"Take a look at the waveform_delay.png attachment. If you look closely, the purple waveform trace is leading the green trace by one sample. The audio ADC or the collection software must be causing it. This isn't a problem, it is just interesting. Your files are stereo but except for the delay and the noise floor they look like they are mono. Is this correct? Is the source a mono microphone?"

[HWB: The hardware is having some quirk where the mono microphone input apparently does not allow to be set to mono mode.]

 back to top

  back to HPWREN news

field research
network analysis
  ~ university of california, san diego ~ © 2000 ~