Vocabulift

Vocabulift is a language learning tool. Strata by Vocabulift analyses broadcast and archive video on one millisecond timeline: word-timed transcripts in 43 languages, speaker separation, translation into 35 languages, shot and logo detection, on-screen text, and music rights matching with cue sheet export.

Vocabulift is listed as a language learning tool, and its own site notes that a related product has been on the market since 2023 by Webinson, the Košice software company behind it. The homepage supplied here describes Strata, a media intelligence service from the same team that reads a programme on a single millisecond timeline: what was said and by whom, in which language, what the picture showed, and which music played along with the rights data behind it. The speech layer is the most developed. Transcripts are timed word by word with a confidence value per word, speakers are separated without prior enrolment, and the language of each sentence is tagged so bulletins that switch languages remain readable. Lexicons supplied by the user feed corrections and entities, the model adds people, organisations, places, events, works, products and dates, and topics follow a taxonomy the customer defines. Translation covers 35 languages, with 43 recognised for input, Slovak and English among them. The vision layer describes shots and scene boundaries, classifies shot types (presenter, interview, report, control room), and detects logos, objects, people and actions through a self-hosted vision-language model, each with a probability. On-screen text such as captions, tickers and credits is captured for as long as it stays visible. The rights layer links detected music to recordings, works and parties, validates ISRC, ISWC and IPI check digits, proposes recording-to-work matches with evidence, flags conflicts for human confirmation and produces cue sheets that are exported as CSV, JSON, Excel or XML. It is aimed at broadcasters, archives, media monitoring services and rights societies rather than individual learners. Delivery is self-hosted: the software runs on the vendor's servers in Nuremberg, Germany, under NixOS, or inside the customer's own environment, and no media is sent to an outside AI provider. Everything is reachable through an OpenAPI 3.1 interface with 75 operations, signed webhooks and live progress over server-sent events. Engagement starts with a pilot measured on the customer's own recordings, followed by a licence or a joint development agreement, so neither pricing nor a free tier is published.