Dialogue takes a transcript or the rough SRT your editor spat out and gives back a clean SRT, a WebVTT and a plain transcript — every cue re-split to the line length that surface wants and checked against its reading speed.
Nothing opens in this browser — this one takes a job and does the work on our own graphics cards in Oregon. The free allowance below is how you try it: 120 cues, no card and no payment, on the key you are issued.
It does not listen and it does not guess at time. A plain transcript with no durations in it is refused rather than timed by estimate.
The problem
What comes out of an editor is usually technically a caption file and practically unusable: cues forty characters too long for the player, two speakers in one cue, reading speeds nobody can keep up with, and a format one platform accepts and the next rejects. Fixing a hundred titles by hand is a fortnight, so the captions stay broken and the search index stays empty.
Every timestamp in the output is copied from your own file. Dialogue re-splits text across cues and never moves a time, so a caption that was in sync going in is in sync coming out.
A clean SRT, a WebVTT and a plain transcript out of one job — because the player wants one, the site wants another, and the search index wants the third.
A transcript with no durations is refused, with the reason, before any work runs. Timing text by guess produces captions that drift, and drifting captions are worse than none.
A transcript is words. A caption file is words with times on them. Turning the first into the second requires either listening to the audio — which Dialogue cannot do, there is no speech model on our machines — or estimating from word counts, which produces a file that is right at the start and a second and a half out by the end. So a transcript with no durations is refused, in those words, before anything is charged.
How it works
An SRT or VTT out of your editor, however rough. Long cues, two speakers in one, stray formatting — all of that is what this is for.
Line length and reading speed are not universal: a player that shows two lines of thirty-two characters is a different job from one that shows one of forty-two. You name the surface and the limits come from it.
SRT, WebVTT and plain transcript, with a report of any cue the reading-speed check flagged. The flagged ones are usually genuinely fast dialogue rather than an error, and they are yours to decide on.
What it will never do
There is no speech model on our machines and that is a deliberate limit, not a gap waiting to be filled. It is why Dialogue can promise that your work never comes near us.
Pricing
A cue is one caption — the unit the file is actually made of, counted off your own timestamps rather than off a runtime you type in. A twenty-minute scene is usually two to three hundred cues, so the free 120 is most of a short title and all of a clip.
A short title or several clips, captioned in full.
A working catalogue captioned properly.
Years of uploads brought up to standard.
Re-splitting the same file for a second surface is free, because the model work was already done — only the arithmetic changes. Your licence key is emailed to you within two working days of payment.
Checkout is our own bitcoin server, not a card processor. The units you buy are added to your key: buy from the Buy more box inside the program and they arrive on their own once the payment confirms, or buy here and paste the invoice number from your receipt into that box.
Questions
Because that would mean you sending us your footage, and we are not going to hold your work.
The entire company is built on never touching content. A speech model on our machines would mean a copy of your video on our disks, and no promise about deleting it afterwards is worth as much as not having it.
Get it timed first, in the editor you cut the title in — most will export an SRT with rough times in a couple of minutes.
Dialogue is refusing you to protect the output. A guessed timing looks fine for thirty seconds and is embarrassing by the end of the scene.
To tell you which cues nobody can read in the time they are on screen.
Most flagged cues are real fast dialogue and you will leave them. The useful ones are the cues that are only fast because the splitting was wrong, and those get fixed in the same pass.
No. The re-splitting work runs on Tower 1's own graphics cards in Oregon and the limit checks run in Python beside it.
A key. One line of text, issued to you and shown exactly once — only its hash is kept here, so it cannot be looked up later and we could not tell you what yours is. Every key arrives with a free allowance already on it, which is the tier at the top of the pricing above, and the work starts the first time you send a job with that key on it.
Each job is metered in the unit named in the pricing, and a unit that fails — a file that will not read, a row the model held back — is not charged and is named in the job report rather than silently skipped. When an allowance runs out the next job is refused before any work starts, with the number it used printed on the refusal.
More units are bought in bitcoin. The checkout is a bitcoin server we run ourselves. Buy from the Buy more box inside the program and the units are added to your key on their own as soon as the payment confirms; buy from the plan buttons above and paste the invoice number from your receipt into that same box. Each payment adds its units once, to one key.
Diff the timestamps against your original if you want the promise proved. They will be identical, which is the whole point.