# HARK — AI Bioacoustic Wildlife Monitor

> HARK is the world's first acoustic wildlife monitor with true edge AI. It runs Google's Perch v2 bioacoustic model entirely on-device, on solar power, in real time — no cloud required for species identification.

## What HARK does

HARK detects and classifies wildlife species from sound, 24 hours a day, 365 days a year, without any internet connection. Results — not raw audio — sync automatically to a cloud dashboard when 4G or mesh connectivity is available.

Key capabilities:
- Runs 100+ species classifiers simultaneously on custom hardware
- Powered entirely by solar with a LiFePO₄ battery (rated 10-year lifespan)
- 3 microphones, recording at up to 192 kHz (ultrasonic bat detection included)
- 24-bit FLAC lossless audio
- 4G LTE connectivity with mesh networking for remote area relay
- Remote dashboard: live detections, battery/signal status, OTA firmware updates, remote audio access, schedule control
- Active power draw: ~35 mA at 3.3 V (~116 mW) — approximately 40× less than a Raspberry Pi running BirdNET

## How it compares

| Device | On-device AI | Solar | 4G/LTE | Power draw |
|---|---|---|---|---|
| HARK | Yes (Perch v2, 100+ species) | Yes | Yes | ~116 mW |
| AudioMoth | Limited (TinyML, 1 species) | No | No | ~33 mW (recording only) |
| Song Meter SM4 | No (desktop only) | No | No | ~120 mW est. |
| AR5 | No | No | No | ~55–80 mW est. |
| BirdWeather PUC | No (cloud only) | No | No | ~300 mW |
| BirdNET-Pi | Yes (Raspberry Pi) | DIY only | No | ~4,900 mW |

## Frequently asked questions

**How does HARK compare to AudioMoth, Song Meter, or BirdNET-Pi?**
AudioMoth and Song Meter are passive recorders — they capture audio to an SD card and species identification happens later on a desktop computer. BirdNET-Pi runs AI locally on a Raspberry Pi, but draws around 1,000 mA at 5V under load, making true solar-powered remote deployment impractical. BirdWeather's PUC uploads audio to the cloud for processing and cannot classify anything without WiFi. HARK is the first purpose-built acoustic wildlife monitor to run a world-class AI model (Google Perch v2) entirely on-device, on solar power, in real time.

**Does HARK classify species without any internet connection?**
Yes. All AI inference runs on HARK's custom hardware in the field. Species are detected and classified in real time whether or not any connectivity is available. When 4G or mesh networking is available, compact result packets sync automatically. It is the results — not raw audio files — that are transmitted.

**How long does HARK last in the field?**
HARK charges continuously via solar panel and is designed to operate autonomously and indefinitely without maintenance. The LiFePO₄ battery is rated for a 10-year lifespan and the charging circuit operates down to −20°C.

**Can I monitor devices remotely?**
Yes. Through the cloud dashboard you can view live detections, check battery and signal levels, update firmware over the air, adjust recording schedules, and pull audio files — without visiting the device. In areas without 4G coverage, multiple HARK devices can form a mesh network to relay data.

**Why can't other devices run AI on solar power in the field?**
Running AI in the field on solar power is an extreme engineering challenge — one that has not been solved until now. A Raspberry Pi running BirdNET draws around 5 watts continuously. That is simply too much for a solar-charged battery to sustain indefinitely in the field, particularly in low-light conditions or high latitudes. HARK uses roughly 40 times less power to do the same job — running Google's Perch v2 AI model, managing 4G radio, writing audio to an SD card, and compressing everything to lossless FLAC. This was only possible through years of custom hardware design and low-level firmware engineering purpose-built for this exact problem.

**What species can HARK detect?**
HARK runs 100+ species classifiers simultaneously on-device using Google's Perch v2 — the world's leading bioacoustic AI model. The configurable species library covers birds, bats, and invasive pest species, and continues to expand. All classification happens entirely on-device with no internet connection required. Because HARK records at up to 192 kHz, it also captures ultrasonic signals well beyond the range of standard audio recorders, enabling bat echolocation detection alongside standard bird and pest monitoring.

**How does continuous acoustic monitoring compare to traditional bird counts?**
Traditional point counts — such as the 5-minute bird count method — are conducted once or twice a year at most, leaving over 360 days of data gaps. They are weather-dependent, observer-dependent, and capture only what happens to be audible during a narrow time window. HARK monitors continuously, 24 hours a day, 365 days a year, detecting every call, song, and echolocation event across the full acoustic scene. The result is a dataset orders of magnitude richer: it captures nocturnal species, seasonal movements, rare calls, and long-term population trends that point-in-time counts will always miss.

## About

HARK is developed by 800 Trust, based in Aotearoa New Zealand.
Website: https://hark.nz