tavi
Distributed Bird Sound Identification
Deploy satellite microphones, identify birds with AI, collaborate across teams
Satellites anywhere
An Android phone or a Raspberry Pi becomes a listening station. It sets silence and noise aside itself, keeps recordings while offline and sends them when the connection returns.
BirdNET identification
Cornell's BirdNET names more than 6,500 species from 3-second recordings, at the hub or on the satellite itself, filtered by what lives near the satellite at that time of year.
Teams
People work in teams with four roles (viewer, member, admin, owner). Each team sees only its own satellites and detections, and one hub hosts many teams.
Settings at every level
Hub, team, account and satellite each set their own values where they may. A level sets the default for the ones below and can lock it.
Verification
Teammates confirm or reject uncertain detections by listening, voting and marking the call on the spectrogram. Their answers calibrate the confidence scores.
Alerts
Rules for rare species, first of the season, silent satellites and more, sent by email, Slack, Discord, Telegram, Google Chat or webhooks.
Species pages
Photos, descriptions and traits for every species heard, with bird names in 38 languages.
Map, analytics and reports
Where and when each species was heard, biodiversity indices, weather beside the detections, and PDF reports.
Privacy by design
Positions leave the hub rounded to about 1 km, IP addresses leave the activity log after 12 months, and each person can export or delete what the hub holds about them.
Latest releases
Badges in these pages
The level a setting or rule applies at, top down: the whole hub, one team, one person, one phone or Pi.
The lowest team role that may do it (viewer, member, admin, owner), or a hub admin.
Where it runs or shows: the web app, the phone, the Raspberry Pi, the hub server.
The release it arrived in; opens that entry of the changelog.