EMSN

Type minstens twee letters. Resultaten verschijnen vanzelf, geen Enter nodig.

NL
Back to Listening
§ S · Struikgewas
Acoustics · low

Struikgewas hears what
sounds low.

The microphones hang on the shed in the garden, sheltered from the rain under the gutter. At shrub and garden height they pick up what moves close by: dunnocks slipping through the bushes, wrens that suddenly belt out a loud song from nowhere, robins marking their territory. But these microphones also catch what passes overhead pretty well.

About the confidence score

An algorithm is no ornithologist

Every detection comes with a confidence score. 92% chance it's a blackbird, 23% chance it's a noctule. The number looks scientific, but it remains a model's guess. Often very good. Sometimes completely off.

A few from this garden. A bicycle bell once became a bullfinch. A bouncing ball down the street, an eider. One pipistrelle sounds almost identical to another, as far as BatDetect2 is concerned. The algorithm misfires there regularly, even at 95%. Rare species are extra suspect. The model has seen fewer of them, so a detection like that stands on shaky ground.

I check the less common and rare species myself. For each detection I play the recording back and see if it holds up. But that leaves me running behind sometimes. I might only get to it days later, and until then a doubtful case just sits on the site.

So those numbers are a hint, not an identification. An algorithm that listens day and night hears more than I ever could on my own, and that's already something.

Latest detection
Tree pipit 77%
★ Hoogtepunthelderste opname die hier is geregistreerd

Groenling

15 mei · 10:13 · 100%

↗ permalink
§ S-02 · Live in the shrubs From the microphones under the gutter in the garden

Just in.

Every detection from BirdNET-Pi berging shows up here straight away. The microphone hangs low. So what you see here is mostly callers and begging youngsters, no fly-overs.

connecting
  1. Listening.

    No detections since you opened this page. A quiet stretch means nothing was heard, not that the pipeline is down.

§ S-03 · Today in view What we've heard so far

The day so far

grouped by species
  1. Song thrush 6x until 04:46
  2. Grey heron 12x until 02:38
  3. Eurasian teal 1x
  4. Dunlin 1x
  5. European robin 1x
  6. Tree pipit 1x
§ S-04 · What are we actually hearing Vocalisation types, top 10

Birds don't just sing. They call, alarm and make contact sounds.

Custom CNN models decide per detection whether it's song, call, alarm or something else. Counts over the last seven days, for the top 10 most-heard species.

Species song call alarm other
Common swift 20,387 10,268 12 5,615
Common wood pigeon 390 1,290 357 194
Eurasian blue tit 200 1,050 335 452
Eurasian magpie 367 333 542 198
Western jackdaw 228 622 7 113
Great tit 224 269 158 142
Common blackbird 9 300 385 33
Common chaffinch 60 280 175 127
Grey heron 2 256 18 32
European greenfinch 8 88 11 22

Song = territorial (song / subsong / display song). Call = communication (call / begging call / flight call / nocturnal migration call). Alarm = predator spotted. Other = raw drumming or unclassified by the model.

§ S-05 · Listening Top 3 per species, top 9 on confidence

What there was to hear low down.

The low shrub layer mostly captures callers and begging youngsters. A script picks the three clearest recordings of the past year per species, every night.

  • Groenling

    Chloris chloris
    • 01 15 mei 26 · 10:13 100%
      alarm ↗ link
    • 02 15 mei 26 · 10:13 100%
      alarm ↗ link
    • 03 15 mei 26 · 10:13 100%
      alarm ↗ link
  • Pimpelmees

    Cyanistes caeruleus
    • 01 2 apr 26 · 08:14 100%
      alarm ↗ link
    • 02 29 mrt 26 · 15:19 100%
      ↗ link
    • 03 21 mrt 26 · 06:50 100%
      bedelroep ↗ link
  • Gierzwaluw

    Apus apus
    • 01 17 jul 26 · 23:21 100%
      roep ↗ link
    • 02 16 jul 26 · 23:50 100%
      vluchtroep ↗ link
    • 03 24 jun 26 · 15:24 100%
      vluchtroep ↗ link
  • Kolgans

    Anser albifrons
    • 01 25 feb 26 · 20:39 100%
      vluchtroep ↗ link
    • 02 22 mrt 26 · 21:58 100%
      vluchtroep ↗ link
    • 03 10 mrt 26 · 02:39 100%
      nachttrekroep ↗ link
  • Appelvink

    Coccothraustes coccothraustes
    • 01 4 mrt 26 · 11:23 100%
      bedelroep ↗ link
    • 02 4 mrt 26 · 11:23 99%
      bedelroep ↗ link
    • 03 10 mei 26 · 10:44 99%
      ↗ link
  • Houtduif

    Columba palumbus
    • 01 14 apr 26 · 08:45 100%
      roep ↗ link
    • 02 27 apr 26 · 21:56 100%
      subzang ↗ link
    • 03 12 apr 26 · 09:15 100%
      zang ↗ link
  • Nijlgans

    Alopochen aegyptiaca
    • 01 26 apr 26 · 08:43 100%
      roep ↗ link
    • 02 26 apr 26 · 08:43 100%
      roep ↗ link
    • 03 26 apr 26 · 08:43 99%
      roep ↗ link
  • Koolmees

    Parus major
    • 01 2 mrt 26 · 09:41 100%
      roep ↗ link
    • 02 14 feb 26 · 18:42 100%
      subzang ↗ link
    • 03 14 feb 26 · 18:48 100%
      alarm ↗ link
  • Merel

    Turdus merula
    • 01 26 apr 26 · 21:23 100%
      roep ↗ link
    • 02 26 apr 26 · 21:19 100%
      zang ↗ link
    • 03 8 mrt 26 · 17:21 100%
      subzang ↗ link

9 soorten · 27 fragmenten getoond · nog 88 soorten in het archief. Selectie gemaakt op 18 jul 26.

Alle soorten bekijken →
§ S-06 · Highlights · all time Hand-picked, with my own context

The classics.

Rarities, one-off passers-by, and moments I don't want to forget myself. Every single one fished out of the archive by hand.

No highlights marked yet. Ronny regularly picks new ones.

§ S-07 · Today versus normal What stands out?
  • Today 6 different species, with a total of 22 detections.
  • Peak today around 02:00 (12 detections). The 30-day average peaks around 21:00.
  • Today fewer detections than the average over the last thirty days (-100%).
Detections per hour 0–23 h
today 30-day average
§ S-08 · Under the hood How Struikgewas is built

Struikgewas runs on a Raspberry Pi 5 (8GB) in a weatherproof enclosure on the shed in the garden. Two microphones hang under the gutter of the shed, sheltered from the rain, at shrub height. The Steinberg UR44 audio interface handles the input. BirdNET-Pi analyses every three seconds, and then the same CNN models as on Bladerdak determine the vocalisation type (song, call, alarm or contact). North/south direction detection doesn't work here yet. That'll come later.