TTAL scripts in and out, dialog-gated loudness QC, C2PA-signed audio, AAF sessions, checksummed delivery packages. The paperwork of a professional dub, produced by the platform.
Resumable multipart ingest goes direct to storage — presigned part URLs, individually retryable parts, masters up to 20 GB. Every file lands with MD5 and SHA-256 checksums, in private storage served only through expiring signed URLs, with every access written to the audit log.
Netflix's TTAL is the script format of the dubbing pipeline, so it is ours too. Import a studio-authored TTAL and unknown fields survive the round trip. Export pre-recording and as-recorded scripts as TTAL JSON or W3C DAPT (TTML2), with SMPTE timecodes on every event.
{
"meta": {
"version": "V2.0",
"type": "DUB_SCRIPT",
"title": "Meridian — Ep 102",
"lang": "de",
"originalLanguage": "en",
"fps": "FPS_2397",
"dropFrame": false,
"timeFormat": "FRAMES"
},
"events": {
"evt_00042": {
"start": 11599,
"end": 11625,
"startSec": 483.774958,
"endSec": 484.859375,
"startTimecode": "00:08:03:07",
"endTimecode": "00:08:04:09",
"txt": "Wir hatten eine Abmachung.",
"transcriptionText": "We had a deal.",
"characterId": "character_0"
}
},
"characters": {
"character_0": { "name": "DETECTIVE SULLIVAN" }
}
}Actors record against a digital rythmoband — the line scrolling to a fixed sync bar, frame-locked to picture — with a segment queue, pre-roll beeps, and uncompressed 48 kHz WAV capture. Takes are versioned, reviewable, and drop straight into the mix. And every cloned voice is gated on a verifiable consent record.
Transcript of the rythmoband demo dialog:
The project frame rate is probed from the master — 23.976 through 29.97 drop-frame — and every timing in the studio is SMPTE, down to the frame. Smart Timing picks the wording that fits each on-screen moment, and full-runtime sync-drift QC catches accumulated offset before a streamer does.
Max drift 12 ms across the full runtime · 214 segments
Loudness is measured the way the delivery spec measures it: BS.1770 dialog-gated integrated loudness, 4× oversampled true peak, loudness range. Netflix −27 LKFS, ATSC A/85 −24 — pass, warn, or fail against the profile you're delivering to, in a report you can hand over.
Netflix (dialog-gated)
Target −27 LKFS ± 0.5 · ceiling −2 dBTP
Every synthesized line carries an inaudible acoustic watermark; every export, a C2PA manifest naming the project, the model, and the consent record behind the voice. Private storage, per-project IP allowlists, and a full media-access audit log — built for TPN-style security reviews.
Consent record
verifiedconsent_ref: esign_5f2a · signed 2026-05-14
Acoustic watermark
verifiedscheme: audioseal · inaudible · present ✓
C2PA manifest
verifiedsignature valid · asserts project · model · consent_ref
sealed into every export · verification API returns the full chain
Printmaster and per-speaker stems at 48 kHz/24-bit with BWF bext and iXML metadata, an ADM BWAV, an AAF session with clips at program timecode, the as-recorded TTAL, and the QC report — assembled under one package ID with a checksummed manifest. An HMAC-signed webhook tells your systems the moment it is ready.
We export AAF, not .ptx — the closed Avid format no third party writes reliably.
voice features preserved, measured against objective benchmarks
languages, cross-lingual from a single voice sample
speech-overlap IoU between source and dubbed timelines, reported per job
Upload a video, get back a dubbed one — translation, cloning, timing, and mix as one pipeline.
ExploreTimbre, prosody, and accent reproduced at a fidelity we benchmark and publish.
ExploreThe entire stack on your infrastructure, or embedded in your product via SDK and API.
ExploreSend us a master and a dialogue list. You get back a checksummed package — printmaster, stems, scripts, QC report — the way your MAM expects it.