Voxtext Translation Manifest Delivery

An optional, off-by-default power-user feature in Voxtext.

Most people will never need this page, and that’s fine. Voxtext’s job is still “drag a file in, get a transcript out.” Translation Manifest Delivery is for the small group of people who already have an automated translation workflow and want Voxtext to hand it work, without anyone copying and pasting between apps.

What it does

When the feature is on, Voxtext does everything it normally does (transcribes locally with Whisper, writes your transcript files) and then also writes a couple of small JSON files, called manifests, that describe the transcript in a form an automation can read:

• One per-video manifest, saved next to each source file. It holds the segment timings, the segment text pre-formatted for translation, the target language, and the folder the finished translation should land in.
• One job manifest per run, saved in a folder you choose. It lists the per-video manifests from that run. Because it’s the only file that lands in that folder, it works as the “go” signal for a watch-folder automation.

Your workflow takes it from there. Voxtext does not translate anything in this mode, and it doesn’t talk to your workflow directly. It just prepares the work and drops the files where you told it to.

Voxtext transcribes locally
        |
        v
video01.manifest.json  (next to the video)
job manifest           (in your watched folder)  --->  your automation notices it,
                                                       translates the segments,
                                                       writes video01.fr.vtt back

Who it’s for

• Teams that produce captions in more than one language and already have (or are building) an automated flow to translate them.
• Anyone comfortable with a tool like Power Automate, Zapier, n8n, or a script that watches a folder.

A word on privacy

Voxtext’s promise is that your audio never leaves your computer, and that stays true here. But a manifest contains the transcript text, and you’re pointing Voxtext at a folder that (in the typical setup) syncs to the cloud. So with this feature on, your transcript text travels wherever that folder syncs to, and whatever reads it (your translation service) sees it. Only turn it on if that’s what you want.

Turning it on

1. Open Options and find the Translation Manifest Delivery card.
2. Tick Enable Translation Manifest Delivery.
3. Fill in the three fields below.

The last two exist because the folder name your computer uses for a synced library often doesn’t line up neatly with the real SharePoint path, and guessing wrong breaks the handoff. So Voxtext asks you for both ends and does the conversion for everything underneath. If you leave them blank, Voxtext still writes the manifests but leaves out sourceFolder (more on that below) and tells you in the status bar.

Once the feature is enabled, a Translation Manifest panel appears on the main screen. Untick the feature in Options and the panel disappears again, no restart needed.

Using it

1. In the Translation Manifest panel, tick Generate Translation Manifest.
2. Pick the target language.
3. Optionally fill in Course / batch name (used to name the job manifest; defaults to the source folder’s name) and a Context hint (see below).
4. Transcribe as usual. Single files and batches both work.

You don’t need to tick VTT yourself. Every manifest points at a <name>_transcript.vtt file, so Voxtext writes one automatically whenever manifests are on, on top of whatever formats you picked.

Context hint. A short note like compliance training video or casual podcast, first names only. It nudges the translator toward the right register and terminology. It’s added as a single line at the top of the text your workflow sends to its translator, and it isn’t part of the translated output.

The files

Naming and location

Voxtext deletes and recreates each manifest instead of overwriting it. Cloud folders often treat an overwrite as a modification rather than a new file, and many triggers only fire on new files. Delete-then-recreate means your automation gets a fresh “file created” event every time.

Per-video manifest fields

Job manifest fields

A machine-readable JSON Schema for the per-video manifest is available alongside this page as manifest.schema.json.

Example

A per-video manifest:

{
  "courseName": "Onboarding Basics - Module 3",
  "sourceLanguage": "en",
  "targetLanguage": "fr",
  "sourceFolder": "/Shared Documents/Training/Onboarding/Module 3",
  "sourceFileName": "01 - Welcome_transcript.vtt",
  "translatePrompt": "Content context: internal training video\n\n[[1]] Welcome to the team.\n[[2]] In this module we'll cover your first week.",
  "segments": [
    { "id": 1, "start": "00:00:00.000", "end": "00:00:02.800" },
    { "id": 2, "start": "00:00:02.800", "end": "00:00:06.350" }
  ]
}

The matching job manifest:

{
  "courseName": "Onboarding Basics - Module 3",
  "sourceFolder": "/Shared Documents/Training/Onboarding/Module 3",
  "manifestFiles": [
    "01 - Welcome.manifest.json",
    "02 - Your First Week.manifest.json"
  ]
}

What your workflow needs to do

Voxtext is only half of the handshake. If you’re building the other half, these four rules keep translated captions from drifting out of sync with the video:

1. One segment in, one segment out. Never merge, split, add, or drop segments. The number of translated segments must equal the number in segments, matched by id.
2. Never renumber ids. [[7]] in means id 7 out.
3. Take timing from the manifest, not the translator. Copy start and end from segments exactly as written. The translation step should only touch the words after each [[id]] tag.
4. Send only translatePrompt to the translator. Nothing else in the manifest needs translating.

Then a typical flow is: trigger on a new .job.manifest.json file, loop over manifestFiles, read each per-video manifest, translate translatePrompt, check that the returned ids exactly match segments, build a WEBVTT file from the timings plus the translated text, and save it into sourceFolder (for example as <video name>.<language>.vtt). If the ids don’t match, stop and flag it rather than writing a broken caption file. A captions file that’s silently out of sync is worse than no captions file.

Tips and gotchas

• sourceFolder must be a SharePoint path, never a local one. A Windows or Mac path like C:\Users\... will make SharePoint-based flows fail. That’s what the two base-folder settings are for.
• Nothing happens if the feature is off. With the master checkbox off, Voxtext behaves exactly as it always has.
• A manifest problem never sinks your transcript. If a manifest can’t be written, Voxtext keeps your normal transcript files and tells you what went wrong.
• Check the status bar. If the job manifest went out without a sourceFolder, Voxtext says so rather than failing silently.
• Tidy up as you go. Each run leaves manifest files behind. Your workflow can move processed ones into a subfolder so the source folders and watched folder stay tidy. Just let your team know what those files are.

What’s supported

This feature is deliberately kept out of the everyday app, and it’s supported on a best-effort basis. It’s built and tested against a SharePoint/OneDrive folder setup feeding a Power Automate flow. The manifest format is plain JSON, so you’re welcome to build against it with other tools, but I can’t promise to troubleshoot your particular setup.

You can download the schema at the following link: MANIFEST_SCHEMA

If something in the manifest format doesn’t work for your workflow, I’d like to hear about it.