Multilingual Transcription Pipeline for a Broadcast Organization

01 — CHALLENGE

A broadcast organization with a large multilingual media library needed accurate transcription and subtitling at a scale no manual team could sustain — including live transcription and translation for major events.

02 — APPROACH

We engineered a two-stage pipeline: high-quality acoustic transcription followed by LLM-based correction using domain vocabulary, then extended the system to batch subtitle correction across the archive and live transcription/translation workflows for streamed events.

03 — DELIVERABLES

• Two-stage STT + LLM-correction transcription pipeline
• Batch subtitle generation and correction workflow
• Live-event transcription and translation configuration
• Domain vocabulary and correction rule sets

04 — OUTCOME

The organization processes hours of multilingual audio in minutes, with correction quality tuned to its own vocabulary — and can caption live events that previously went untranscribed.