Free · no ads
Do the noise.
Badly. Competitively.
Everyone hears the same two-second sound. Everyone tries to be it. Then the server says, to the point, how close each of you got. No acting talent required — the people who win are usually the ones least embarrassed.
Start a room
Two to ten people, in the same room or on a call. Everyone hears the clip at once and performs at once — nobody waits for a turn.
How a round goes
- Everyone hears the clip once. That's the only time you get it clean, so listen properly.
- Everyone records at the same time, alone, hearing nobody else. No waiting for a turn.
- The moment the booth shuts, everybody's take plays back with a score out of 100 on it.
- Throw a potion at whoever sounded dangerous. Then save the whole session as one clip.
Eight kinds of noise
Every round deals one sound from the kinds your room picked. You hear it once, then everybody goes at the same time.
- CreaturesThings with a throat.
- MachinesThings with a speaker.
- CartoonPhysics that never happened.
- VoicesSomebody said something.
- ReactionsThe noise a person makes involuntarily.
- SpaceNothing out here should be audible.
- SportsCrowds, whistles, and regret.
- SongsNow actually sing it.
Your character
Everyone in the room is somebody, and they mouth along to their own take in the video you save at the end. Colours and faces are free; hats, props and auras come in packs.
Worth sending to people
At the end of a match you can save the whole session as one video — every take, every potion, every score, cut together. Vertical for shorts, or landscape for everywhere else.
Questions people ask
- What is a voice imitation game?
- Everyone hears the same short sound — an animal, a machine, a shout — and everyone tries to reproduce it with their own voice at the same time. Each attempt is scored on how close it got, and then everybody listens back to how badly it went. PartyDub's version runs in a browser for two to ten people.
- Do I need to download or install anything?
- No. Mimic runs in the browser on a phone or a laptop. You need a microphone and a link; there is nothing to install and no app store involved.
- How is an imitation scored?
- By measurement, not by a model. Your recording is compared to the original on four things: where your attacks landed relative to the clip's, how your pitch moved, how your loudness rose and fell, and your rhythm. The result is one number out of a hundred, and it appears the instant the recording window shuts because nothing has to be asked of anything.
- Does it matter whether my voice is deep or high?
- No, and this is deliberate. The pitch comparison removes the median from both contours before comparing them, so what is measured is the SHAPE of the movement rather than the notes. A bass voice and a child's voice performing the same rise and fall score identically. Loudness works the same way: a quiet microphone is not a worse score.
- Why did I get a low score when I thought I nailed it?
- Usually timing. The results screen splits it into two numbers because they are different mistakes: being consistently late by the same amount is barely audible and costs very little, while being scattered — early on one attack, late on the next — is what actually sounds wrong and is what the score reflects. The breakdown tells you which of the two happened.
- Can we save a round to share?
- Yes. At the end of a match you can export the whole session as one video — every player's attempt, every thrown potion and every score, cut together — as a vertical clip for shorts or a landscape one for everywhere else. It is rendered in your browser; the recordings are not kept on a server afterwards.
How it's judged
One number out of a hundred, and no AI anywhere near it. The same measurements the ranked dubbing mode uses — how your pitch moved, where your attacks landed, how your loudness rose and fell — compared against the original's. It doesn't care how deep your voice is or how loud your microphone was, so a big voice and a small one can both score full marks on the same clip. And because nothing has to be asked of a model, the score is just there the instant you stop.