TAN, minus the jargon

The technical version lives in the repository. This is the plain one.

Why I built this

Movies are mixed for theaters. Big quiet-to-loud swings feel great in a theater. At home you just end up watching the whole thing holding the remote. I have to turn up the volume for conversations and down for explosions. I like watching movies, so I thought I would try and solve this problem once and for all.

What it does

TAN keeps a hand on the volume knob so you don't have to. Whispers get nudged up. Explosions get eased down. Everything else is left alone. The remote stays on the couch.

How it catches an explosion before you hear it

It cheats time a little. TAN holds the sound back 8 milliseconds and uses the head start to peek at what's coming. That's under a hundredth of a second; you can't perceive it. The bang shows up already turned down.

Files get a better deal. TAN reads the whole file before touching anything, so it knows about every explosion in advance. Every adjustment lands smooth.

Why not make everything the same volume

Because that sounds awful. Flattening the whole mix kills the movie. TAN figures out the show's own normal loudness and pulls everything gently toward it. Dialogue comes up to normal. Explosions settle just above normal. Your volume knob keeps meaning what it always meant.

It also measures loudness the way ears actually work. A deep rumble and a clear voice can carry identical numbers and sound completely different. TAN uses the same hearing model the streaming industry standardized on, so its idea of too loud matches yours.

Hasn't Dolby done this

Mostly. Their version needs loudness data baked into the movie at the studio, and it only runs on hardware that licenses their tech. That's a walled garden. I hate walled gardens.

TAN does the job by listening. No metadata, no licensed hardware. The code is public, free, and readable.

What's under the hood

Rust, a language that's fast and catches whole categories of bugs before the program runs. Good qualities for something that processes sound all day without glitching.

WebAssembly lets the same code run in a browser. That's why the demo page can normalize your own files, or a playing YouTube tab, without uploading anything anywhere.

Where it's headed

Today it handles files and browser audio. Next it should sit inside your computer, phone, or TV and fix everything you play automatically. After that I want it recognizing dialogue specifically, so voices stay clear when the music tries to bury them.