How fast is Studio?
On a laptop with an Apple M3 chip, Studio runs BirdNET 2.4 on an hour of audio in 13 to 17 seconds, about 210 to 275 times faster than real time. A season of 1,000 hours of recordings takes roughly 4 to 5 hours on the same laptop.
Same results as BirdNET-Analyzer
Studio runs BirdNET, the model made by the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology and Chemnitz University of Technology. It builds BirdNET's spectrograms itself and runs the model through ONNX Runtime, so we checked it against BirdNET-Analyzer 2.4.0, the BirdNET team's own program: on the same recording, place and week, both found the same 10 detections at the same times, with scores within 0.005 of each other.
What was measured
- Computer: a Mac with an Apple M3 chip (8 cores) and 8 GB memory, macOS 27.2.
- Recording: one test recording, one hour, mono, 48 kHz, 16-bit WAV (346 MB), already in the computer's file cache.
- Studio's engine, built from the Studio source on 8 November 2026 and run from its command line, timed from start to finish including loading the model. Two runs: 17 s, then 13 s.
Limits of this test
- One computer and one recording. Windows, Linux and slower laptops are still to be measured; results will be added here.
- Many short files carry more start-up cost per hour of audio than one long file.
- Perch 2.0 is a larger model and runs slower; it has not been timed here.
Repeat it yourself
# Wildlife Network Studio engine (from github.com/wlnw-org/wlnw-studio) cargo build --release -p wlnw-engine --bin wlnw wlnw scan test.wlnw <folder of recordings> wlnw place test.wlnw 1 <latitude> <longitude> <UTC offset> wlnw use test.wlnw birdnet-2.4 time wlnw run test.wlnw <models folder>
Measured on 8 November 2026. Timed it on your own computer? Send us the numbers and we will add them.