Third-party tests · independent validation

We don't ask you to take our word.
We show you the papers.

Three independent tests by three separate teams. Same conclusion every time.

BENCHMARK 01 · THIRD-PARTY TESTING

Independent validation — 98.7% F1 accuracy.

01 · WILDER SENSING · THIRD-PARTY · F1 ACCURACY 02 · JAN DRACHMANN · FIELD · 5 DETECTORS · 11 SPECIES

98.7%

F1 ACCURACY

Geoff Carss

Geoff Carss & Annabel Jeffries

Wilder Sensing

"We tested a range of existing solutions using real acoustic data, and BioSonic clearly stood out, achieving 98.7% accuracy (F1 score) and giving us full confidence in the results we're bringing to the platform."
F1 ACCURACY 80 85 90 95 100 Kaleidoscope 84.2% BioSonic 98.7%

BioSonic's AI model uses Image recognition on Spectrogram. This leads to a step change in accuracy.

Metric comparison — Precision, Recall, F1 EXPAND ↓

higher is better · bars start at 50% baseline

PRECISION · BioSonic +22.6pt

BioSonic
98.3%
Kaleidoscope
75.7%

RECALL · BioSonic +7.9pt

BioSonic
99.6%
Kaleidoscope
91.7%

F1 SCORE · BioSonic +15.9pt

BioSonic
98.9%
Kaleidoscope
83.0%

BIOSONIC · Confusion Matrix · Accuracy 99.7%

PREDICTED BATPREDICTED NO BAT
ACTUAL BAT730
TRUE POSITIVE
3
FALSE NEGATIVE
ACTUAL NO BAT13
FALSE POSITIVE
4,579
TRUE NEGATIVE

KALEIDOSCOPE PRO · Confusion Matrix · Accuracy 95.8%

PREDICTED BATPREDICTED NO BAT
ACTUAL BAT543
TRUE POSITIVE
49
FALSE NEGATIVE
ACTUAL NO BAT174
FALSE POSITIVE
4,503
TRUE NEGATIVE

FALSE POSITIVES (Wrong "bat" calls)

Cost: investigator time wasted reviewing non-bat audio, inflated species counts

BIOSONIC

13

KALEIDOSCOPE

174

Kaleidoscope: 13.4× more

FALSE NEGATIVES (Missed bats)

Cost: protected species overlooked, incomplete EIA, compliance risk

BIOSONIC

3

KALEIDOSCOPE

49

Kaleidoscope: 16.3× more

Same data, very different outcomes. BioSonic delivers 98.9% F1 with only 16 errors across 5,325 predictions. Kaleidoscope Pro misses nearly 9% of true bats and produces 13× more false calls.

BENCHMARK 02 · SOUTHERN DENMARK

Manual vs BioSonic.

Detector ranking

Identical

D4 > D5 > D2 > D1 > D3 — same order whether counted by hand or by AI. Same ecological conclusion about which sites had the most activity.

Detections per detector

same order, slightly higher counts
D4
JD
23,527
BS
25,514
D5
JD
2,370
BS
3,505
D2
JD
1,156
BS
1,688
D1
JD
391
BS
461
D3
JD
94
BS
118
JD — Manual BIOSONIC

11 → 100

NATTER'S BAT CALLS

Jan found 11. BioSonic found 100.

Going through tens of thousands of files by hand, it's easy to miss a few. On review, every one of BioSonic's 100 Natter's bat calls was validated as correct. That's the payoff of AI on large datasets — rare species don't slip through.

Species breakdown — Common & Rare species comparison EXPAND ↓

Common Species

BioSonic detects more Pipistrel (+203%) and Trold (+16%)

Manual BioSonic
Dværg
13,556
10,669
Trold
11,433
13,248
Pipistrel
2,257
6,834

Rare & Protected Species

Separate scale — these are the ones that matter for Annex II compliance

Manual BioSonic
Vand
112
221
Frynse
11
100
Dam
0
0
Brun
40
41
Syd
55
67

Weather integration

Jan also found interesting bat behaviours compared to wind speed and rain with BioSonic's automatic weather integration.

Bat activity vs wind speed

And the human cost

2TB bat data, comparing BioSonic vs Kaleidoscope.

According to Consultancy in Southern England

KALEIDOSCOPE WORKFLOW

490 hrs

12 weeks to client

vs

BIOSONIC WORKFLOW

163 hrs

4 weeks to client

✓ 327 hours back · This means £21,255 saved at standard UK consultant rate

Watch

BioSonic on BatAbility Club with Neil Middleton.

Josef Carlson

Run it on your own data

A 30-minute demo.
Bring a WAV, leave with the report.

josef.carlson@biosonic.se · +46 72 744 65 85