Guide · AudioMoth recordings
Yes, AI can sort AudioMoth bat recordings by species. For European bats the realistic options are BioSonic, a browser tool that takes AudioMoth WAV files as recorded and scored 98.7% F1 in a third-party benchmark by Wilder Sensing; the BTO Acoustic Pipeline, a low-cost desktop app with cloud processing that covers all European bats; and Kaleidoscope Pro, licensed desktop software with UK and European classifiers. Researchers who code can also use the free BatDetect2 (trained on UK data, non-commercial licence only) or Tadarida, the open-source engine behind France's Vigie-Chiro programme, and SonoChiro is commercial software from Biotope. Whichever you choose, record at 384 kHz and check low-confidence and Myotis calls by hand.
BioSonic accepts AudioMoth WAV files as recorded, at any file length, plus WAV and FLAC from other full-spectrum detectors and Anabat zero-crossing files.
Before you deploy
AudioMoth is a low-cost, open-source recorder from Open Acoustic Devices. It records uncompressed WAV files to a microSD card at 8,000 to 384,000 samples per second and keeps time with an onboard clock set to UTC.8 Out of the box it records at 48 kHz and medium gain,13 which is set up for audible sound, not bats. These are the settings to change in the AudioMoth Configuration App before a bat deployment.
| Setting | For bats | Why |
|---|---|---|
| Sample rate | 384 kHz | The app offers 8, 16, 32, 48, 96, 192, 250 and 384 kHz.9 A recording holds frequencies up to half its sample rate, so 384 kHz reaches 192 kHz and 250 kHz reaches 125 kHz. European bat calls span roughly 15 to 120 kHz.24 The BTO recommends 384 kHz for its Pipeline because, without an anti-aliasing filter, lower rates cause clipping problems.14 The 48 kHz default only reaches 24 kHz. |
| Gain | Start at medium and test | Five levels: low, low-medium, medium, medium-high and high, with medium as the firmware default.10 Make a test recording at the site and check that loud passes do not clip and faint calls still show. |
| Recording mode | Triggered for most surveys | Triggered recording starts on an amplitude threshold or, from firmware 1.8.0, a frequency trigger. A minimum trigger duration keeps recording after the last trigger, so a quieter feeding buzz after a bat call is not lost.11 Continuous recording keeps every second but makes far bigger files (see below). |
| Filters | Only below your lowest-calling species | Low-pass, band-pass and high-pass filters are applied to the audio before it is written to the card.11 Open Acoustic Devices' bat example uses a 50 kHz high-pass filter, but in Europe that would remove most of the noctule, serotine and northern bat group (about 18 to 35 kHz) and the common pipistrelle's peak near 45 kHz.24 If you filter, set the cut-off below the lowest species you expect. |
| Schedule | Dusk to dawn | The Configuration App can set recording periods that follow sunrise and sunset, and estimates battery life for the settings you choose.26 |
| Time and file names | Set the clock before every deployment | Files are named by date and time, for example 20260601_213000.WAV, or …T.WAV for triggered recordings, with an optional device ID prefix and daily folders.10 Settings survive a battery change but the clock resets, so set the time again.13 The BTO recommends recording in local time so results show the correct time.14 Whichever you choose, write it in your method. |
| Metadata | Keep files as recorded | Each WAV header carries a comment with the recording time, gain and battery state, and current firmware also writes a GUANO metadata block with make, model, serial number, firmware version and timestamp.10 Renaming or re-saving files in an audio editor can lose this. |
| Microphone and housing | Do not seal it in a plastic bag | The BTO notes that AudioMoth's microphone does not perform as well as a commercial bat detector at higher frequencies, often losing the weaker ends of calls that matter for Myotis and horseshoe bats, and that self-noise can reduce classifier performance. A plastic bag makes the loss worse.14 |
AudioMoth writes 16-bit mono WAV, so every sample takes 2 bytes.12 That makes continuous file sizes easy to work out. Triggered volumes depend on bat activity, noise and your threshold, so the figure below is our storage calculator's mid-range estimate, not a measurement.18
| Mode | Per hour | Ten-hour night | 300 nights |
|---|---|---|---|
| Continuous, 384 kHz | about 2.8 GB | about 28 GB | about 8.3 TB |
| Continuous, 250 kHz | 1.8 GB | 18 GB | 5.4 TB |
| Triggered (estimate) | varies | about 1.8 GB | about 540 GB |
A ten-hour continuous night at 384 kHz nearly fills a 32 GB card. Triggered recordings go into a T.WAV file that stores only the triggered audio plus the length of each silent gap. Any WAV program can open it, and the AudioMoth Configuration App can expand it back to real time and split it into shorter files if a tool needs that.12
Options compared
Six tools come up for European bat work. Each is a serious option, and the right one depends on where you survey, how much data you have, whether your results can be shared and whether you can write code. We make BioSonic, so we have put it first; every other entry is described from its maker's own pages, with numbered sources at the bottom.
| Tool | What it is | Region and species | Install or web | Cost model | AudioMoth files | Good for |
|---|---|---|---|---|---|---|
| BioSonic (Biosonic AB, Sweden) | AI platform that filters noise, suggests a species with a confidence score for each bat recording, flags feeding buzzes and social calls, and logs every check. 98.7% F1 in a third-party benchmark by Wilder Sensing.1 | European bat species | Web browser, nothing to install | Commercial: you buy a package of analysis gigabytes; free trial | WAV as recorded, any file length21 | Teams with many detector-nights who want ID, checking and reporting in one place |
| Kaleidoscope Pro (Wildlife Acoustics) | Desktop analysis software with bat Auto-ID classifiers. Wildlife Acoustics says to always verify a suggested ID manually.3 | United States, Canada, UK, Europe, South Africa and the Neotropics3 | Desktop install | Commercial, single-user licence | Reads full-spectrum WAV; we found no AudioMoth-specific statement on the maker's pages | Offline processing and work on several continents |
| BTO Acoustic Pipeline (British Trust for Ornithology) | Classification service: a free desktop app, processing in the cloud and results on a management dashboard.2 BTO reports a 93 to 100% true positive rate for European bats from its own checks.17 | All European bats, with classifiers for every European country, plus North Africa and the South Caucasus; also small mammals, bush-crickets and moths2 | Desktop app plus cloud | Free allowance of credits if you share results; opting out uses more credits; extra credits in bundles; bespoke plans for large-scale or commercial work16 | Named in BTO guidance: 384 kHz, firmware 1.2.2 or later for date and time file names. Files longer than 5 seconds are split into 5-second chunks.14,15 | Low cost when results can be shared; other taxa from the same recordings |
| BatDetect2 (Mac Aodha and colleagues) | Open deep-learning model that finds each echolocation call and predicts its species. The authors report a mean average precision of 0.88 on a 17-species UK dataset.25 | Default model trained on UK data4 | Install with Python; command line | Free under CC BY-NC 4.0: non-commercial use only4 | Runs on a folder of .wav files4 | Researchers who code, on non-commercial projects. The authors strongly recommend validating on recordings of known species first. |
| Tadarida and Vigie-Chiro (MNHN, France) | Tadarida is an open-source toolbox for detecting and classifying animal sounds.5 It identifies the recordings of Vigie-Chiro, the citizen-science bat monitoring programme run by the Muséum national d'Histoire naturelle since 2006, where participants check results on a web portal.6 | Built for France (Vigie-Chiro) | Toolbox needs technical set-up; Vigie-Chiro runs through its web portal | Open source (LGPL-3.0)5 | We found no official statement on AudioMoth; check the Vigie-Chiro protocol | Researchers and Vigie-Chiro participants in France |
| SonoChiro (Biotope, France) | Commercial software that finds and classifies bat calls; version 4 adds SonoView for checking sonograms. Biotope says it cannot entirely replace manual identification.7 | 111 species, with regional classifiers such as Mediterranean, Northern Temperate and British Isles7 | PC install; online licence check every 30 days | Commercial, single-user licence | Reads WAV and batcorder RAW; AudioMoth is not named7 | France and the Mediterranean; desktop work |
BioSonic runs in the browser with nothing to install. You upload WAV, FLAC or Anabat files of any length, and the whole team reviews the results together in shared projects. Recordings are stored in the EU, mainly in Sweden, and a new AI model is trained and deployed every weekend. You buy a package of analysis gigabytes; the bigger the package, the lower the price per GB, and recordings are kept for 9 years. Ask us for a quote, or start with a free trial on your own AudioMoth nights. Two honest limits: BioSonic needs an internet connection, and it covers European species only.
No single published test has run all six tools on the same AudioMoth recordings. The fairest check is to run a few nights you have already identified through two or three tools on your shortlist and compare. For tool-by-tool detail, see our Kaleidoscope Pro alternatives guide and the comparisons with Kaleidoscope Pro, the BTO Acoustic Pipeline, BatDetect2, Tadarida and Vigie-Chiro and SonoChiro.
Workflow
The steps below are for BioSonic. The order is the same in any tool: keep the files intact, let the software do the sorting, and spend your own time on the calls that need a human.
Copy every file from each SD card into one folder per site and deployment. Keep the file names and do not convert, rename or re-save the files, so the date, time and metadata stay intact.
Drag the folders into one BioSonic project in the browser. Any file length is accepted, so long continuous recordings do not need cutting into short files first.
Noise files are filtered out, and every bat recording gets a species suggestion with a confidence score. Up to 7 species can be found in one recording, and feeding buzzes and social calls are flagged.
Work through the low-confidence identifications and the look-alike groups on the spectrogram, confirming or correcting each one. Then export species, date and time per recording to CSV or Excel.
Myotis calls are hard for every classifier. A 2024 test of three classifiers in the north-eastern United States, Kaleidoscope Pro among them, found none could reliably identify Myotis to species,23 and the BTO warns that AudioMoth's microphone often loses the weaker ends of calls that matter for Myotis and horseshoe bats.14 Check every Myotis pass by hand and report to genus (Myotis sp.) where you cannot confirm the species. More broadly, Bat Conservation Trust guidance says automated classifications should be manually verified, ideally at least one recording of each species on each survey date, and especially notable records.22
The BTO asks for full-spectrum WAV recorded in real time, does not support zero-crossing files, and splits files longer than 5 seconds into 5-second chunks.15 For AudioMoth it recommends firmware 1.2.2 or later, so file names carry the date and time the Pipeline needs, a 384 kHz sample rate and local time.14
Whatever tool you use, results are only as good as the recordings, so it pays to check one test night from each new site before you deploy for a whole season.
Time saved
This example uses only figures already published on this site: our time calculator, built from Wharton's customer data (0.245 analyst hours per GB for a traditional Kaleidoscope-class workflow and 0.0815 hours per GB with BioSonic),19 and the storage calculator's estimate of about 1.8 GB a night for a triggered AudioMoth.18
| Step | Calculation | Result |
|---|---|---|
| Detector-nights | 10 AudioMoths for 30 nights each | 300 |
| Data volume | 300 nights at 1.8 GB each | about 540 GB |
| Traditional workflow | 540 GB at 0.245 hours per GB | about 132 hours (3.8 weeks at 35 hours) |
| With BioSonic | 540 GB at 0.0815 hours per GB | about 44 hours (1.3 weeks) |
| Analyst time saved | 540 GB at 0.1635 hours per GB | about 88 hours |
The rates match the single night Wharton tested side by side: 4,200 files (10.3 GB), around 3,000 of them noise, took about 2 hours 30 minutes in Kaleidoscope Pro and about 50 minutes in BioSonic, 66% less time, most of it from automatic noise filtering.20 These rates come from Wharton's own data, not from AudioMoth recordings, and continuous AudioMoth recordings hold much more empty audio per GB, so the per-GB rate may not carry over to them. Put your own volumes into the time calculator, or upload one night you have already analysed and compare.
FAQ
Yes. AudioMoth records full-spectrum WAV files at up to 384 kHz, which bat classifiers can read.8 For European bats you can upload them to BioSonic in the browser, process them with the BTO Acoustic Pipeline desktop app, or run them through Kaleidoscope Pro, BatDetect2, Tadarida or SonoChiro. The software filters out noise and suggests a species for each bat recording, but you still need to check low-confidence calls and the hard groups, Myotis above all, by hand. The BTO also notes that AudioMoth's microphone often loses the weaker ends of calls, which matters most for Myotis and horseshoe bats.14
384 kHz, the highest rate AudioMoth offers. A recording holds frequencies up to half its sample rate, so 384 kHz reaches 192 kHz, well above the roughly 15 to 120 kHz range of European bat calls.24 The BTO recommends 384 kHz for its Acoustic Pipeline because, without an anti-aliasing filter, lower sample rates cause clipping problems.14 250 kHz still reaches 125 kHz and makes smaller files, but test it on your own species first. Do not deploy for bats at the 48 kHz default, which only reaches 24 kHz.13
Yes, with limits. The BTO Acoustic Pipeline app is free to download, and processing is free within an allowance of credits if you share your results with the Pipeline.16 BatDetect2 is free, with a default model trained on UK data, but its CC BY-NC 4.0 licence covers non-commercial use only, so it is not for paid consultancy work.4 Tadarida is open source under LGPL-3.0 but needs technical set-up.5 BioSonic offers a free trial on your own recordings and is otherwise a paid service.
Yes. Upload AudioMoth WAV files as recorded, with no conversion, renaming or splitting. BioSonic accepts any file length, so long continuous recordings can go in whole.21 It runs in the browser with nothing to install, filters out noise files, suggests a species with a confidence score for each bat recording, and lets the whole team check results together before exporting to CSV or Excel. Recordings are stored in the EU, mainly in Sweden.
Triggered, for most bat surveys. Continuous recording at 384 kHz writes about 28 GB in a ten-hour night, and most of it contains no bats. A triggered AudioMoth starts on an amplitude threshold or, from firmware 1.8.0, a frequency trigger, and writes the triggered audio to a T.WAV file that keeps the timing of the gaps, so it can be expanded later in the AudioMoth Configuration App.11,12 Set a minimum trigger duration so quieter sounds just after a call, such as feeding buzzes, are kept. Choose continuous only when you need an unbroken record of the night.
Not reliably, with any tool. A 2024 test of three classifiers in the north-eastern United States found none could reliably identify Myotis calls to species,23 and the BTO notes that AudioMoth's microphone often loses the weaker ends of calls that matter for identifying Myotis and horseshoe bats.14 Check every Myotis pass on the spectrogram and report to genus (Myotis sp.) where you cannot confirm the species. Bat Conservation Trust guidance also expects automated identifications to be checked by hand, ideally at least one recording of each species on each survey date.22
Sources
Figures about BioSonic come from pages on this site. AudioMoth settings and other tools are cited to their own makers and publishers.
Last verified: 2026-11-18.
Upload a night of AudioMoth recordings and compare the species list, and the time it took, with your current workflow. Free trial, in the browser, nothing to install.
support@biosonic.se · +46 72 744 65 85