Audio description services that stand up to an audit.

Listed by

The American Council of the Blind lists our work

The ACB’s Audio Description Project keeps the directory that blind and low-vision viewers actually use to find described films. We have thirty-two described films; the directory lists thirteen of them so far, and every new title goes to them as it is published.

We are one of twenty-five listed providers covering some 24,000 described titles — the same list as Netflix, Prime Video, Disney+, Max, Hulu, Paramount+, Peacock, Apple TV and PBS. We are on it because the work qualified, and every title we add goes to them too.

See our titles in the ACB directory — their site, their filter, not a page we control.

We are a listed provider, not a partner or an affiliate of the ACB, and we claim no endorsement from them. They list what meets their criteria.

From a production we described

“Mike was such a pleasure to work with. He responded quickly, explained the process thoroughly, and delivered on time. Thank you for really listening to our needs and providing an affordable service.”

We produced the audio description for the Normal Isn’t Real trailer.

The legal part, plainly

Does the law actually require this?

Does WCAG 2.1 Level AA require audio description?

Yes. Success Criterion 1.2.5 Audio Description (Prerecorded) is Level AA. At Level A you may substitute a text alternative for the video. At Level AA you may not, so a real description track is required rather than a transcript.

When is the deadline for government video?

The DOJ rule at 28 CFR 35.200 adopts WCAG 2.1 Level AA for state and local government under ADA Title II. Public entities serving 50,000 or more: 26 April 2027. Smaller entities and special districts: 26 April 2028. Title II binds government, not private business — if you are a private company, your obligation comes from Title III case law, your own contracts, or a customer requirement, and it is worth knowing which.

Does every video need it?

No, and we would rather say so before you pay. 1.2.5 applies where the picture carries information the soundtrack does not already convey. If your narrator already says what is on screen, the video may conform as it stands. Send us the file and we will tell you if you need nothing.

What about Section 508?

Federal is not a future deadline. It is already past. The 2017 refresh of Section 508 (29 U.S.C. 794d) adopted WCAG 2.0 Level A and AA for electronic content, with a compliance date of 18 January 2018. WCAG 2.0 Level AA already contains SC 1.2.5, so federal agencies, and the contractors delivering video to them, have been required to provide audio description for prerecorded video since then.

The practical difference is only the version. Section 508 points at WCAG 2.0 AA; ADA Title II points at 2.1 AA. For audio description and captioning the requirement is materially the same, and work produced to the 2.1 standard satisfies both. If you are delivering under a federal contract, ask us for the conformance report against 508 specifically and that is what you will get.

This is a plain reading of published rules, not legal advice, and we are not lawyers. Anything that turns on your specific obligation should go past your own counsel.

Who is actually doing this

There is a real room behind the studio tier, and it has a name

The studio-voice reads are recorded at The Hitbox, our own commercial recording studio, open in its commercial location for over eight years. Albums recorded and produced in it, and thousands of songs mixed in it, long before any of them were films.

The audio on every tier is directed by an engineer who has been recording and mixing since 2001, working professionally in studios since 2003, with twenty years of mastering behind that. That is the difference between using a tool and being sold by one: on the synthetic tiers the machine does the reading, and a person who has spent a career on level, placement and intelligibility decides whether what came back is acceptable. Most of the corrections we make are not ones a novice would know to make.

Picture restoration is digital. We do color, grain, damage and stabilization work on scans, and audio restoration on the tracks that come with them. Nobody here touches film, and this page will never imply a photochemical lab we do not have.

Check it with a network tab

No third-party anything

There is no analytics script on this site, no tag manager, no advertising pixel and no embedded font service. No cookie is set by visiting this page. Loading it contacts no server but ours.

That is not a privacy posture we adopted for a page — it is why the Content-Security-Policy can say script-src 'self' with no exceptions, and it is one fewer thing for your own procurement review to write up.

Read this part

What we will not claim

The conformance report verifies placement, coverage, pace, loudness, track structure and flags. It then states plainly that it does not establish whether the description is accurate — whether it says what is on screen, names the right person and omits nothing a sighted viewer relies on. Measurement can refute a cue and cannot confirm one; establishing accuracy still takes a person watching the film against the script.

The sign-off lines are left blank rather than pre-filled. A report that rubber-stamps is worth nothing to an auditor.

Two passes look at the picture, and they are not the same pass. An AI model reads the whole runtime at two frames a second, and every motion at five, before a word of description is written, with a coverage check that refuses a gap longer than two minutes. Then an object detector runs over the film and every cue is cross-checked against what it saw — which catches the error a first draft makes, a cue saying “he” over a shot of two women. A person reviews the finished script and hears every mix before it ships.

We state the limit of that plainly, because it is the question worth asking of anyone selling this. A detection is strong evidence; a non-detection is not. The model is trained on modern color photographs and under-detects on an old black-and-white print — and it knows nothing about a hearse or a gantry. So a cue that contradicts what was detected is a defect we fix; a cue the detector cannot confirm is a question we ask, never a failure we report. A tool that treated absence as fact would be confidently wrong.

Your print never leaves our machines. The detector runs locally on a six-megabyte model, not against a hosted vision API, so nothing is uploaded to a third party to be indexed — which for an archive holding an unreleased element is usually the first thing asked and the last thing answered.

Our catalog also links out to Tubi, Pluto and YouTube, and those play in their players — we make no conformance claim about software we do not control.

Who you are contracting with