A song reaches microphones at different times. Research published on 9 May 2026 combines those delays with AI bird-song detection to locate the source.[1]
GPS-synchronised recorders supply the timing. Comparing matching sounds produces candidate positions; inconsistent estimates are rejected.[1]
In a test with nine recorders, eight speaker positions and songs from six species, 99% of retained locations fell within five metres of the true source. Mean error was 1.14 metres. But only 34% of four-second windows containing at least one second of song yielded a position. No sounds from the edge speaker were successfully located.[1]
A separate deployment used 61 recorders across roughly 6.25 hectares of Pennsylvania forest over four days in May 2022. Its maps resembled the spatial patterns from human surveys; individual wild-bird positions were not directly validated.[1]
The result offers ecologists coordinates for selected songs. A blank area on the map can still contain a singing bird.[1]
Sources & further reading
- A fully automated framework for acoustic identification and localization of terrestrial wildlife at scale Communications Biology · 9 May 2026 · Accessed 20 Oct 2026