Alexandrina Council · Community Safety

Turn hours of recordings into one clear answer.

Dog Bark Detector listens to your WAV recordings, finds every bark using Google's YAMNet AI, and hands back a timestamped Excel report, individual MP3 clips, and prompts ready to paste into Copilot — all processed on the council PC, nothing uploaded anywhere.

02:14:36 — bark detected, 87% confidence

What it actually does

No cloud service, no subscription, no guesswork. Point it at a folder of recordings and it does the listening for you.

Built for evidence, not just numbers

Every bark becomes a case-ready record

Each detection gets a timestamp, a duration, a confidence score, and its own MP3 clip you can listen back to and attach directly to a complaint file. The Excel report groups activity into clusters, flags quiet periods, and surfaces the patterns an officer actually needs — longest bark, busiest recording, time-of-day spread.

Runs on the council PC

Python, ffmpeg and the YAMNet model install directly from python.org and GitHub — no winget, no admin portal, no recordings ever leave the machine.

Writes the first draft for you

Seven built-in prompts turn the same data into a formal notice, a plain-English update for the complainant, or a case-file entry — paste into Copilot and go.


The time this actually saves

Based on real-world testing, running a full week of continuous recordings through Dog Bark Detector takes about 5 hours on average — compared with the weeks it would take to listen through the same audio by ear.

Listening manually
21 working days
Dog Bark Detector
~5 hours
24 hours x 7 days = 168 hours of continuous recordings Dog Bark Detector processes the full week in about 5 hours on average Listening manually, at an 8-hour work day: 168 hours / 8 hours = 21 working days (over 4 working weeks)

In practice: a full week of continuous recordings that would take a staff member the better part of a month to get through by ear is processed in well under a single business day — and it's unattended machine time, not hands-on staff time, so nobody has to sit and listen to any of it.

This speeds up listening, not judgement. Results are meant to be checked, not taken on faith — YAMNet is a strong classifier but not infallible, and it can occasionally miss a very faint bark or misclassify another sound. Every detection comes with its own MP3 clip specifically so a person can quickly confirm what was flagged. Always spot-check the results before they go into a formal notice, evidence file, or any other official record.

Ready to see it running?

Six guided steps from a folder of recordings to a finished report.

Walk through the steps