AccessCast

Audio description · Captions · Easy read · हिन्दी · Italiano · Español · Français · Português

One video. Every viewer.

Try it with sound on

Close your eyes.

Listen to the same clip twice. First the original sound. Then the version for someone who can't see the screen: a voice says what is on screen, in the pauses between words.

What a blind viewer hears today.

The same clip, with description in the pauses.

Someone speaks Description plays A pause where description fits

Press play, then close your eyes.

Who's watching?

Pick who it's for. The player opens in that version, and you can mix them.

Six steps, one video

How it works?

  1. Who does it: speech recognition

    We find every spoken word and the moment it is said. That shows where the pauses are.

  2. Who does it: Gemma 4

    Gemma looks at frames from each scene and lists only what it can see, including text on screen.

  3. Who does it: code

    Code picks which pauses get a description and how many words fit. If a pause is too short, the video can stop for a moment.

  4. Who does it: Gemma 4

    Gemma writes each line to fit its pause. It also writes the easy-read captions and the Hindi, Italian, Spanish, French and Portuguese versions.

  5. Who does it: code, then Gemma repairs

    Code checks every line: timing, length, names nobody said, guesses about feelings or jobs. Gemma fixes what fails, and code checks it again.

  6. Who does it: a person

    A person approves what the public hears. Anything not yet approved is marked as a draft.

Try the checks

Gemma proposes. Code decides.

The scene

Loading the example…

What we measured

Results.

Every number here is read from a file our scripts wrote. Where an experiment has not run yet, it says so.

Access is the law: jump to the dates

The law already asks.

Description and captions are no longer extras. These rules say when they must be there.