Audiomoth Configuration testing results

AudioMoth is a general purpose device that has been used in a wide variety of applications worldwide from monitoring poachers gunfire, to tracking crickets. The basic branch of the firmware is highly configurable and reflects this general approach (at the time of writing version 1.4.4).

Hence some degree of experimentation is needed to find the best settings that work for UK bats. The testing process we followed is summarised on this page, with the full results in this spreadsheet. Please review this Audiomoth whitepaper for explanation of the features being tested.

If you have any feedback on this testing, or your own Audiomoth findings, then please do get in touch.

SUMMARY of Testing results & Outcomes

Sample Rate

Numbers of bats detected by AutoID in recordings using (near) continual recording was tested with sample rates of 192, 256 and 394 kHz. These recording were also manually IDed for verification. Summary below:

  • Result – Number of bats detected using sample rate of 192 and 256 kHz is broadly similar, while 394kHz records 10-20% less bats.
  • Outcome – Use 256kHz. As we are in a Lesser Horseshoe (LHS) area we standardised on 256kHz. In areas without LHS, it would be sensible to use 192kHz to reduce filesize.

Gain

Using all 5 gain settings the number of bats Auto ID detected in recordings using (near) continual recording was tested. Summary below:

  • Result – High gain was the worst, with Low gain second worst. The three medium settings (Low/Med, Med & Med/High) were similar with Medium usually slightly the best of the three.
  • OutcomeUse Medium Gain

High Pass

Using a range of frequencies for high pass filter, the numbers of bats Auto ID detected in recordings using (near) continual recording was tested using Med and High gain. The aim was to reduce the Audiomoth internal noise recorded by finding highest, ideal, setting for high pass that still allows reliable detection of Noctules at 20kHz. Summary below.

  • Result – This “ideal” setting of High pass is dependent on Gain
  • Outcomes – For Medium gain use 25kHz, for High Gain use up to 45kHz.

AMPLITUDE THRESHOLD

For each setting of gain, a range of Amplitude thresholds were used. The total size of the files recorded was measured, and the number of bats detected in these recordings was noted.

Effect of gain and amplitude threshold on file size

Overall by far the the biggest effect on file size is background noise. A windy night in trees, or heavy rain will mean the Audiomoth records continually whatever the setting of amplitude threshold. It was also noted that even on quiet nights there can be smaller variances between individual Audiomoths. This may be due to variability in internal noise of the device.

  • Summary Results
    • The setting of gain has a large effect on the “optimal” setting of Amplitude threshold. Or put another way, a amplitude threshold that produces tiny files at low gain, will result in continual recording at high gain.
    • The internal noise of the Audiomoth will result in continual recording at low settings of amplitude threshold, until a point is reached where small changes in amplitude threshold rapidly decrease the file size until about 10-15% of the size of continual recording (CR). After this, the effect of further increases in amplitude threshold diminishes.
Effect of file size on number of bats detected

The number of bats detected in every recording set of gain vs. amplitude threshold was analysed. Summary below:

  • Results – The number of bats detected remains broadly constant until the total file size is approx 10% that of continual recording (CR). As the total file size falls below 10%CR the number of bats detected gradully reduces.

The truncated files were also expended using the Audiomoth config. app to their full 100% CR size, and bat AutoId reapplied to the expanded files.

  • Results – In all cases, the expanded files resulted in fewer bats autoIDed than original truncated files. While this may be due to elimination of some false positives, manual ID verification of a small sample showed this was not always the case.
Summary of Amplitude Threshold testing

The aim of this testing had been to produce recommendations as to the ideal settings of amplitude threshold for detecting UK bats. We have not been able to do this. The main problems encountered were:

  • Variability of the night time background noise, which can change rapidly between nights or even within a night.
  • The smaller variability between the internal noise of individual Audiomoths.
  • The sensitivity of both the gain and amplitude threshold on the resulting file sizes mean there is a small “Goldilocks” band in which “small and good” bat recording can be made.
  • We have found no way to predict this “Goldilocks” setting. The AM team are clearly aware of this, as in their whitepaper they recommend sampling in the field using the red flashing LED to try and establish the right setting. But this is very difficult as it would need to be done for every device in every location, and more critically it would not allow for noise variability within the night time recording period.

Outcome of Amplitude Threshold testing

For UK bat recordings the only way we can see to ensure reliable recordings is not to use amplitude threshold, or set it at very conservative setting that may frequently result in near continual recordings.

Duration testing

For the Somerset landscape scale surveys we plan to use rechargeable batteries. We are currently testing how many nights Audiomoth can record for in various configurations.

Future work

Experimental Triggering Firmware

There is also an experimental branch of Audiomoth firmware developed in 2019 by the team at Southampton working with Adrian Bicker (versions 1.0.9 and 1.0.10). This is hard-coded for southern UK bats and uses ultrasound triggering rather than a Amplitude Threshold.

We are doing some comparison testing of this triggering firmware vs. amplitude threshold.

There are also discussions ongoing about progressing this triggering firmware, to make it configurable.

FUTURE AUDIOMOTH TESTING PLANNED

We still have open questions and plan the testing below. However if anyone has already tested these, or knows of others test results then please do get in contact.

  1. Triggering 1.0.9 vs Amp.Threshold 1.4.4 Firmware – Using the “best settings” from testing to see which detects the most bats.
  2. Case – Is a wire mesh vent, or the porelle vent of the official case better at letting bat sound pass to the AudioMoth?