honk & whistle

A sound, your voice, a score.

Warming up the sound engine — about 17 MB on the first visit, instant after that.

honk & whistle

Listen, mimic, score, share.

Pick a sound and tap Play.
No attempt yet
About

What is this?

honk & whistle is a tiny voice-mimicry game. Pick a sound, listen, then try to make the same sound with your voice. You get a score from 0 to 10 and a sharable link to challenge someone else to beat it.

How the score works

Two things are measured and blended:

A perfect mimic of a sound against itself scores 10. A genuine attempt usually lands in the 5–9 range. Boing and Hum are the friendliest references; Bell is hardest because no human voice can match a pure 880 Hz harmonic.

Privacy & security

Audio never leaves your device. The microphone opens only for the 3-second recording window and closes immediately. There is no server: the whole game runs in your browser. No accounts, no cookies, no analytics, no tracking.

The only network request after the page loads is the one-time fetch of the audio model from Google's TF Hub (about 17 MB). Nothing about you, your voice, or your scores is uploaded — anywhere, ever.

On a phone?

Designed for mobile. Open in Safari (iOS) or Chrome (Android), tap a button, allow microphone access. If the iPhone silent switch is on, iOS Safari will mute the reference sound — flick it off. The first visit downloads the audio engine; afterwards it's cached and instant.

Anonymous play data

Once you complete a day's three rounds, we send your three scores and the puzzle date to our server so we can understand whether the scoring is fair. Nothing else — no audio, no IP, no fingerprint. See the privacy page for the full detail.

Get in touch

Feedback, ideas, or want to know more? Email thegreatsuperbobo@gmail.com.

Keyboard: Tab between controls, Space or Enter to activate.

Having trouble?

1. Can't hear the sound?

2. The score seems wrong?

3. Still stuck?

Email thegreatsuperbobo@gmail.com with what's happening — your device, browser, and what you tried.

Testing

For verifying the scoring pipeline. Self-test scores the current reference against itself — should always be ~10. If it isn't, something on this device is preventing the model from running correctly.