Illustration: a felt landscape on the coast. Sound from a video screen becomes a waveform, the waveform becomes subtitle tiles, one line of tiles turns into a translation, a camera iris marks objects in the picture and a record carries a rights tag.

Strata · multilingual media intelligence

Video in.
Timeline out.

Open to partnerships in audiovisual intelligence

Strata puts what a programme contains on one timeline: what was said and by whom, in which language, what was on screen, and the music rights data behind it. One self-hosted service with an open API, built and run by Webinson, the team behind Vocabulift.

What it does

Three areas.
One timeline.

Broadcasters, archives, media monitoring services and rights societies ask the same questions of a programme: what was said, what was shown, what music was played and who is owed for it. Strata answers all three on a shared clock, so each answer can be checked against the others.

Live · 43 languages

Speech and language

What was said, by whom and in which language, word by word.

Live

  • Transcripts timed to the word, each word with its confidence
  • Speakers: who spoke when, without enrolment
  • The language of every sentence, for programmes that switch
  • Corrections and entities from your lexicons; people, organisations, places, events, works, products and dates from the model
  • Topics from your own taxonomy, translation into 35 languages, a summary and keywords
Live

Video understanding

The picture, read on the same timeline as the speech.

Live

  • Shot boundaries and scenes, and what kind of shot each is: presenter, interview, report, control room
  • Logos, objects, people and actions from a self-hosted vision-language model, each with the model's own probability; brands you name are checked in every frame
  • Text on screen (captions, tickers, credits) for as long as it stays: Greek and English today, other scripts by adding their data
  • Events from your own systems, such as audio identification, playout or schedules, on the same timeline, with the sentences, speakers, text, shots, entities and topics around each
Live

Rights data

Detected music connected to recordings, works, rightsholders and identifiers.

Live

  • Recordings, works and parties with ISRC, ISWC and IPI, their check digits validated
  • Recording-to-work matching across Greek and Latin script, with the evidence for every proposal
  • Conflicting records found, then confirmed or dismissed by people, with an audit trail
  • Cue sheets generated from what aired or imported from CSV and Excel; corrected, approved and exported as CSV, JSON, Excel or XML

Real output

One programme.
Every layer.

A 44-second Greek news clip, analysed with a newsroom profile: a lexicon of place and organisation names, a news taxonomy, translation into English and on-screen text in Greek and English. Below is its first sentence with everything else Strata found at that moment, then the song it played, and then the picture of a launch broadcast, read second by second.

0.54 – 2.13 s · speaker S1

Καλησπέρα σας από την Αθήνα.

  • enGreetings from Athens.
  • whoSpeaker S1, one of three in the clip
  • entityΑθήνα: a place, from the lexicon
  • screenΕΙΔΗΣΕΙΣ 20:00 · Αθήνα Ζωντανή μετάδοση
  • topicsEuropean affairs 0.97 · Media and technology 0.77
  • shotShot 1 of 5, scene 1 of 2

Strata's demo run of 2 October 2026: eleven stages in about 20 seconds on a laptop. Every layer keeps the start and end of what it describes, in milliseconds.

// GET /v1/assets/ast_…/timeline?from_ms=540&to_ms=2130, trimmed
{
  "items": [
    { "layer": "speech.sentences",
      "start_ms": 540, "end_ms": 2130,
      "data": { "text": "Καλησπέρα σας από την Αθήνα.",
        "language": "el", "speaker": "S1",
        "confidence": 0.968 } },
    { "layer": "speech.translation.en",
      "start_ms": 540, "end_ms": 2130,
      "data": { "text": "Greetings from Athens.",
        "origin": "model" } },
    { "layer": "speech.entities",
      "start_ms": 1640, "end_ms": 2130,
      "data": { "text": "Αθήνα", "kind": "place",
        "source": "vocabulary", "confidence": 1.0 } },
    { "layer": "vision.text",
      "start_ms": 0, "end_ms": 10250,
      "data": { "text": "Αθήνα Ζωντανή μετάδοση",
        "confidence": 0.92 } },
    { "layer": "speech.topics",
      "start_ms": 540, "end_ms": 9770,
      "data": { "categories": [
        { "label": "European affairs",
          "confidence": 0.966 } ] } }
  ]
}
0:34 / Heard

A song is identified

An audio identification system reports To fos tis polis by Anna Demeti, ISRC GRX9D2600001, in Latin letters. Strata takes it in as an event on the clip's timeline.

0:34 / Matched

The catalogue knows it

The ISRC leads to the catalogue's recording «Το φως της πόλης», written in Greek, and the on-screen credit «Το φως της πόλης Άννα Δεμέτη» lines up with it.

After / Reported

The cue sheet writes itself

The bulletin's cue sheet is generated from the match, reviewed, approved and exported for the rights society, with every change kept in its history.

Launch coverage · 1:25 – 2:44

The picture, read like the soundtrack

  • 1:25New shot: a report from the launch pad
  • 1:25Logo: Axiom Space · 0.97
  • 1:25Rocket 1.00 · tower 0.95
  • 1:30Action: smoke rising · 0.97
  • 1:39Text on screen: LIFTOFF · 0.97
  • 2:30New shot from space: Earth 0.99, spacecraft 0.98
  • 2:43Grid fins · 0.93

NASA's public-domain coverage of a crewed launch, read by Strata on a laptop with an 8-billion-parameter vision model. Each confidence is the model's probability of yes against no; what is not in the picture scores near 0.

// GET /v1/assets/ast_…/detections, trimmed
[
  { "timestamp": "00:01:25.875", "type": "shot",
    "label": "hard" },
  { "timestamp": "00:01:25.875", "type": "logo",
    "label": "Axiom Space", "confidence": 0.971 },
  { "timestamp": "00:01:25.875", "type": "object",
    "label": "rocket", "confidence": 0.997 },
  { "timestamp": "00:01:30.775", "type": "action",
    "label": "smoke rising", "confidence": 0.969 },
  { "timestamp": "00:01:39.275", "type": "text",
    "label": "LIFTOFF", "confidence": 0.967 },
  { "timestamp": "00:02:43.900", "type": "object",
    "label": "grid fins", "confidence": 0.931 }
]

Open API

Built to be
built on.

Strata is an HTTP service first: 75 operations, described in OpenAPI 3.1 and browsable at vocabulift.com/docs. A console shows the same media, rights data and cue sheets to people; the API connects your own systems.

01 / Ingest

Files or links

Upload a programme or give its address; Strata fetches it, checks it and probes it. Retries are safe with an Idempotency-Key.

  • Multipart
  • URL
  • Idempotent
02 / Analyse

Profiles and jobs

Queue an analysis with a stored profile or a configuration of its own, and follow its stages live as server-sent events.

  • Profiles
  • Jobs
  • SSE
03 / Read

Layers and exports

Each layer on its own, the merged timeline, what the picture shows as one timed list, transcripts in the original or a translation, and exports as VTT, SRT, text, CSV and JSON.

  • Timeline
  • Transcripts
  • Exports
04 / React

Signed webhooks

Notifications when media is ready, a job ends, a match is proposed, a conflict opens or a cue sheet moves on, signed the Standard Webhooks way.

  • HMAC
  • Retries
  • RFC 9457

Five calls

From a file to a timeline

  • 1POST /v1/assets: upload the programme
  • 2POST /v1/jobs: analyse it with a profile
  • 3GET /v1/jobs/{id}/events: follow it live
  • 4GET /v1/assets/{id}/timeline: read every layer in one place
  • 5GET /v1/assets/{id}/detections: what the picture shows, timed

Every key has its scopes; every error is an RFC 9457 problem whose pointer names the field. See the reference, the OpenAPI document and the error types, or sign in to the console with a key.

# 1. upload
curl -H "Authorization: Bearer $KEY" \
     -F [email protected] -F name="Bulletin 20:00" \
     https://vocabulift.com/v1/assets

# 2. analyse it with a profile, safely retried
curl -H "Authorization: Bearer $KEY" \
     -H "Idempotency-Key: bulletin-2026-10-02" \
     -H "Content-Type: application/json" \
     -d '{"asset_id": "ast_…", "profile_id": "prf_…"}' \
     https://vocabulift.com/v1/jobs

# 3. follow its stages as they happen
curl -N -H "Authorization: Bearer $KEY" \
     https://vocabulift.com/v1/jobs/job_…/events

# 4. read speech, pictures and events together
curl -H "Authorization: Bearer $KEY" \
     https://vocabulift.com/v1/assets/ast_…/timeline

# 5. only logos and objects, the likely ones
curl -H "Authorization: Bearer $KEY" \
     "https://vocabulift.com/v1/assets/ast_…/detections?types=logo,object&min_confidence=0.8"

A pilot, not a promise

Start with your
own material.

Strata runs today. A pilot shows how well it does on your programmes: measured against your references, inside your workflow, before anything is signed.

  1. 01

    Share samples

    A few hours of radio and television, with reference transcripts and rights data where you have them.

  2. 02

    Measure

    Word error rate, timing, speakers, on-screen text and matching precision, with speed and cost per hour, reported on your material rather than ours.

  3. 03

    Integrate

    Through the API and webhooks, or as a private deployment in your environment or a European data centre.

  4. 04

    Decide

    A licence with technical assistance, or joint development under a research and development agreement.

Each party keeps its pre-existing intellectual property. Access rights and ownership of joint results are agreed for each cooperation.

Deployment and data

Private where
it matters.

Media archives and live broadcasts are sensitive. Strata is built so that you decide where processing happens and what is kept.

EU

Hosted in Europe

Strata runs on Webinson's servers in Nuremberg, Germany, configured declaratively with NixOS. This site and its API are among them.

Open

Models on the server

Speech, speakers, entities, topics, translation, on-screen text and the vision model run on the server itself, on models with published weights. No media is sent to an outside AI service.

Private

Or on your hardware

One program, one database file and model files pinned by checksum: Strata runs inside your environment, on CPU or GPU, sized in writing before integration.

GDPR

Your data, your project

We work under a data processing agreement: pilot material is used for your project only, never to train general models, and deleted when it ends.

Built by Webinson

One team.
Product and engineering.

Webinson s.r.o. in Košice, Slovakia, has developed Vocabulift since January 2023 and brought it to market that spring. Strata grows out of the same work on speech and language. The team has 15 years of software engineering behind products like these.

Partnerships

Let's test it
on your material.

We are interested in a commercial agreement with technical assistance and in research and development cooperation. Send us a few samples, ask for an API key, or book half an hour with the team.