Skip to main content
Dryad

buzzfindr: Automating the detection of feeding buzzes in bat echolocation recordings

Cite this dataset

Jameson, Joel (2024). buzzfindr: Automating the detection of feeding buzzes in bat echolocation recordings [Dataset]. Dryad. https://doi.org/10.5061/dryad.v9s4mw72b

Abstract

Quantification of bat communities and habitat heavily rely on non-invasive acoustic bat surveys the scope of which has greatly amplified with advances in remote monitoring technologies. Despite the unprecedented amount of acoustic data being collected, analysis of these data is often limited to simple species classification which provides little information on habitat function. Feeding buzzes, the rapid sequences of echolocation pulses emitted by bats during the terminal phase of prey capture, have historically been used to evaluate foraging habitat quality. Automated identification of feeding buzzes in recordings could benefit conservation by helping identify critical foraging habitat. I tested if detection of feeding buzzes in recordings could be automated with bat recordings from Ontario, Canada. Data were obtained using three different recording devices. The signal detection method involved sequentially scanning narrow frequency bands with the “Bioacoustics” R package signal detection algorithm, and extracting temporal and signal strength parameters from detections. Buzzes were best characterized by the standard deviation of the time between consecutive pulses, the average pulse duration, and the average pulse signal-to-noise ratio. Classification accuracy was highest with artificial neural networks and random forest algorithms. I compared each model’s receiver operating characteristic curves and random forest provided better control over the false-positive rate so it was retained as the final model. When tested on a new dataset, buzzfindr’s overall accuracy was 93.4% (95% CI: 91.5% - 94.9%). Overall accuracy was not affected by recording device type or species frequency group. Automated detection of feeding buzzes will facilitate their integration in the analytical workflow of acoustic bat studies to improve inferences on habitat use and quality.

README: buzzfindr: Automating the detection of feeding buzzes in bat echolocation recordings

Enclosed files consist of recordings used to train and test a classifier for automated detection of feeding buzzes in recordings of bat echolocation. The R script for developing and testing the classifier is also included. The associated manuscript is currently in revision phase.

Description of the data and file structure

The following excerpt from the manuscript in progress described the recordings:

Files for training the classifier-
I compiled data to train the feeding buzz classifier from recordings of bats made at five locations throughout Ontario, Canada, with three different recording devices over three years (see Table 1). The three devices were the Song Meter SM2BAT+ coupled with a SMX-US microphone, the Song Meter SM4BAT-FS coupled with a SMM-U2 microphone, and the Song Meter Mini Bat with an integrated microphone, all manufactured by Wildlife Acoustics Inc. Recordings were made with a 384 kHz sampling frequency. Due to data sharing agreements, the precise locations of the recording sites cannot be provided. Recordings were from four bat species: Silver-haired Bat (Lasionycteris noctivagans), Big Brown Bat (Eptesicus fuscus), Hoary Bat (Lasiurus cinereus), and Little Brown Myotis (Myotis lucifugus). I compiled recordings of feeding buzzes and recordings without feeding buzzes by visually inspecting spectrograms of the recordings with program Audacity 2.4.2 (Boston, MA). Calls were selected irrespective of their signal-to-noise-ratio. I highlighted each observed feeding buzz from the approximate start of the buzz to just after the last perceived call in the buzz sequence and saved the highlighted sequence as a separate file. For recordings without buzzes, the entire file was used.

These are located in the "buzz training data" folder. Within this directory, there are 6 subdirectories that contain the .wav files used to train the classifier from 5 geographic locations. The subdirectories correspond to the 5 following locations identified in the manuscript:

Site 1 = "buzzes_rr"

Site 2 = "buzzes_tb"

Site 3 = "buzzes_o"

Site 4 = "buzzes_sp"

Site 4 = "buzzes_spmylu"

Site 5 = "buzzes_u"

Each of these subfolders (e.g. buzzes*rr, buzzestb etc.) contains a further set of subf*olders ("buzzes", "buzzes.labels", "non_buzzes"). These are described below.

"buzzes": Contains the .wav files of true buzzes used to train the classifier

"non_buzzes": Contains the .wav files of non-buzzes used to train the classifier.

"buzzes.labels": Contains the Audacity files and txt. files created from labelling the locations of buzzes within the .wav files in the "buzzes" folder.

Files for testing the classifier-
I tested two instances of buzzfindr, one with the RF model and one with the ANN model, on a set of 889 recordings from nine new sites across Ontario, Canada (Table 2). Habitat and equipment deployment conditions were identical to those used for the training data. Each third of the test recordings was from a different recording device model and species frequency group (high-frequency / low-frequency) and buzz class (contained a buzz / lacked a buzz) were equally distributed within each device type.

These are located in the "buzz testing data" folder. Within this directory, there are 10 subdirectories. Nine of these contain the .wav files used to train the classifier from 9 new sites. The subdirectories correspond to the 9 new sites identified in the manuscript:

Site 1 = "I"

Site 2 = "RR"

Site 3 = "Ch"

Site 4 = "Sj"

Site 5 = "Sa"

Site 6 = "Sjj"

Site 7 = "Cr"

Site 8 = "E"

Site 9 = "Ci"

The subdirectory called "Compiled*test*files" contains the compiled testing data set from all nine sites. This compiled set was used to more easily test the classifier on a single set of files instead of having to loop through multiple directories.

Each of these subfolders (e.g. buzzes*rr, buzzestb etc.) contains a further set of subf*olders ("buzzes", "non_buzzes"). These are described below.

"buzzes": Contains the .wav files of true buzzes used to train the classifier

"non_buzzes": Contains the .wav files of non-buzzes used to train the classifier.

Analysis scripts-

The R script for training and testing the classifier is also included. These are located in the directory called "Analysis scripts". Within this directory there are two files. The first is called "Buzz Data Extraction and Modelling (March 2024)" and contains the R script for developing and testing the classifier. The second file is called "Det_Param_Combo" and contains the results of the exercise undertaken to identify the best detection parameters to specify for the signal detection algorithm used in the classifier. A description of the methods used in the R function development and testing is provided in the manuscript.

Sharing/Access information

There is no other information. The associated manuscript is currently in submission phase. For additional information, contact the author: joel.jameson@wsp.com

Methods

I compiled data to train the feeding buzz classifier from recordings of bats made at five locations throughout Ontario, Canada, with three different recording devices over three years (see Table 1). Due to data sharing agreements, the precise locations of the recording sites cannot be provided. Recordings were from four bat species: Silver-haired Bat (Lasionycteris noctivagans), Big Brown Bat (Eptesicus fuscus), Hoary Bat (Lasiurus cinereus), and Little Brown Myotis (Myotis lucifugus). I compiled recordings of feeding buzzes and recordings without feeding buzzes by visually inspecting spectrograms of the recordings with program Audacity 2.4.2 (Boston, MA). Calls were selected irrespective of their signal-to-noise-ratio. I highlighted each observed feeding buzz from the approximate start of the buzz to just after the last perceived call in the buzz sequence and saved the highlighted sequence as a separate file. For recordings without buzzes, the entire file was used.

I tested two instances of buzzfindr, one with the RF model and one with the ANN model, on a set of 889 recordings from nine new sites across Ontario, Canada (Table 2). Habitat and equipment deployment conditions were identical to those used for the training data. Each third of the test recordings was from a different recording device model and species frequency group (high-frequency / low-frequency) and buzz class (contained a buzz / lacked a buzz) were equally distributed within each device type.