Mirelo
  • Mirelo SFX
  • Audio-to-MIDI
  • API Documentation
  • Studio
  • Gibberize
  • Audio-to-MIDI
  • MCP
  • Adobe Premiere
  • DaVinci Resolve
  • Roblox
  • Company
  • Studio Docs
  • Blog
  • Pricing
Open Studio
Loading page
Mirelo

Bring your videos to life with synced, spot-on audio — the easy way

Contact Us
XInstagramDiscordLinkedInTikTok

Company

  • About us
  • Careers
  • Support

Legal

  • Security
  • Terms of service
  • Acceptable use policy
  • Privacy policy
  • Cookie policy
  • Imprint
  • Report a concern

Product

  • Go to Studio
  • Mirelo SFX
  • Gibberize
  • Audio-to-MIDI
  • Mirelo MCP
  • Pricing
  • API Docs
  • Studio Docs
  • Changelog

©2026 - Mirelo AI GmbH All rights reserved

Back to blog

By CJ Simon-Gabriel | July 17, 2026

You don't need words to sound convincing: Meet Gibberize

A political speech can feel important before you understand a single sentence. A passionate argument can sound completely plausible even when the language itself means nothing. That's the idea behind Gibberize — though honestly, we didn't come up with it. Our model did. 😊

Mirelo SFX, our video-to-sound model, had a bug. Whenever it saw someone speaking in a video, it would generate speech… sort of. Since the model can’t lip-read (yet) and we never tell it what anyone is saying, it just made up something that sounded like speech. Convincing intonation, plausible rhythm, clear emotions … but zero actual words. Pure gibberish.

Technically, this made perfect sense: people talking on video usually produce speech. Practically, it was a real problem. When you want sound effects, you don't want fake dialogue layered on top. So we got to work, and Mirelo SFX no longer does this. Bug fixed. 🎉

But something kept nagging us. When we showed someone these gibberish videos, they laughed. Almost every time. Once I was meeting with the CEO of a large tech company. He skipped our slides, went straight to the website, and uploaded his own video: four people talking; worst-case input! 🫣 But when the model generated its enthusiastic nonsense, he … laughed! And said: "That's actually pretty cool."

As it kept happening, we just realized: maybe this isn’t a bug. Maybe this is a feature. And thus, the idea for Gibberize was born.

Gibberize takes any video of people talking and dubs it with fluent, expressive, completely meaningless speech.

Sometimes, it may make more sense than the original.

Sometimes, it produces “fake English with a foreign accent”…

… which makes some French people even less intelligible^^ [I’m French, BTW]:

Try it out. Upload a video, let it speak nonsense, and write your own subtitles. You’ll see: how something sounds can be more powerful than what it says.

Try Gibberize →

More from blog

Generate Sound Effects in Claude, ChatGPT and Cursor
July 30, 2026

Generate Sound Effects in Claude, ChatGPT and Cursor

Generate and edit sound effects for your game, app, or website by describing them to Claude, ChatGPT, or Cursor. Mirelo MCP is live now.

Turning audio to MIDI
July 9, 2026

Turning audio to MIDI

Together with Kyutai, we're open-sourcing MuScriptor — a multi-instrument audio-to-MIDI model — and launching a free Audio-to-MIDI tool in Mirelo Studio that transcribes full mixes into separate instrument tracks.