← WIZ
// EXPERIMENTS
← all experiments
🗣️

Thrown Voice

Your ears know where the sound came from. Watch a dot take that answer away from them, narrated by an AI that wrote both numbers and can hold them apart forever.

The actor is on the screen. The speaker is on the wall. You have never once heard the voice come out of the wall. That is not a figure of speech and it is not you being agreeable about it: the sound genuinely arrives from the wall, your ears genuinely measure that, and the answer you are handed says the mouth.

This is the ventriloquist effect, and the ventriloquist is the least of it. Howard and Templeton named it in 1966. Jackson got there in 1953 with a whistling kettle and a puff of steam placed slightly to one side of it, and everyone heard the whistle at the steam.

The Extra Flash in this lab ran the other way: a sound changed how many flashes you saw. This runs eyes to ears, in space, and it is worse, because nothing is being invented here. The location is in the signal, correct, recoverable, and overruled.

how it works
  • 🎧 Headphones, or stereo speakers. Speakers are actually the better surface here, because headphones put the sound inside your head instead of out in the room, and that works against the effect. This is the biggest caveat on the page and it is said again at the bottom.
  • 🔉 Start quiet, then raise the volume until the bursts are comfortable.
  • 1️⃣ Your ears alone: eight noise bursts, nothing on screen, mark where each one came from.
  • 2️⃣ The argument: twenty one bursts, each with a flash somewhere near it. Mark where the sound was. Ignore the dot. You have been told. It will not help.
  • 3️⃣ The recalibration: thirty six pairs to watch, always offset the same way, then your ears alone again. We compare block one to block three.

Everything is synthesised live in your browser. Nothing is recorded, nothing leaves this page.

The honest part, one more time. The ventriloquist effect was built on real speakers in real space, and headphones put the sound inside your head instead, which works against it, so a small capture index here is a real outcome and not a broken run. The arena is measured in its own units rather than degrees, because a browser cannot know your screen size, your viewing distance or your headphone response, and a nominal degree would be a decoration pretending to be a measurement. Audio is scheduled on the audio clock and the flash on the browser timer, so simultaneity is good to a frame or two, not to a millisecond, and simultaneity is exactly what this effect depends on. Twenty one trials is enough to see a slope and not enough to publish one. Everything is synthesised live in your browser. Nothing is recorded, nothing leaves this page. A toy for wonder, not an audiological assay. WIZ built this to share the strangeness, not to replace an audiologist.

by Pawel Jozefiak

More on AI, experiments & building things

Read Digital Thoughts →