Turtle Pond: Wildlife Observation Solutions

Architecture

Mon, 30 Mar 2026 14:57:38 +0000

Architecture

Heket is two main pieces, both written in Python.

Audio Pipeline:

Audio input (RTSP via ffmpeg)
Detection stage (heuristics / thresholds)
Feature extraction
Classification
Storage (simple sqlite database)

Results Web Site

Simple Flask app (not meant for public facing use)

Future versions may incorporate ML-based classification.

Detections

Mon, 30 Mar 2026 14:57:38 +0000

Detections

The current MVP shows:

Last 10 detections
Embedded playback for each event
“Add to training set”
Confidence scoring
Retraining/model switching

Maybe the future:

Species filtering

Example Detection Output

Setup

Mon, 30 Mar 2026 14:57:38 +0000

Setup

To setup and run Heket, you’ll need:

a camera that exposes a RTSP stream
a Linux box to run the pipeline and web site

Instructions for downloading and installing Heket may be found at Heket’s Github repo. There are instructions for both a native Linux install or a Docker install.

Model

Mon, 30 Mar 2026 14:57:38 +0000

Model

In order to categorize frog sounds, Heket needs a model of labeled frog sounds so that new recordings can be accurately classified.

In my particular area, I don’t have many frogs (maybe a half dozen) so the model Heket ships with will not detect all frogs.

I hope that as people adopt Heket that they’ll create and share their frog models.