Invisible Neighbors: An AI Field Notebook for the Sounds Around Us

Description and use

The app lets users record outdoor sounds, run BirdNET locally in the browser for bird identification, and export a report with original recordings and evidence.

Problem

Walkers and beginner birders lack a simple way to capture and analyze ambient sounds to identify nearby wildlife without relying on external services.

Outcome

The application enables offline, privacy‑preserving bird identification by running BirdNET locally in the browser, allowing users to capture, analyze, and export sound passports without external servers or paid APIs.

Workflow

• 1. Record or import a short outdoor audio or video file.
• 2. Open the listening‑stop page to play the media and request analysis.
• 3. Browser downloads BirdNET v2.4 model from Zenodo and runs TensorFlow.js inference in a worker.
• 4. App splits audio into 3‑second clips, analyzes the first 60 seconds, and keeps top five predictions per clip.
• 5. User reviews timestamped evidence, adjusts the minimum score threshold, adds field notes.
• 6. Export a ZIP with original recordings, an HTML report, and structured JSON (sound passport).

Evidence

BirdNET suggests possible birds, and each suggestion links back to the moment in the recording that produced it.

Source

The application has three pages: Notebook, Listening stop, Sound passport.

Source

When someone requests analysis, the browser fetches the official model archive from Zenodo. TensorFlow.js runs inference in a worker, keeping the work separate from the interface.

Source

Export creates a ZIP containing original recordings, a readable HTML report, and structured JSON.

Source