Audio/video transcription: from recording to translated captions

Meetings, interviews, and subtitle projects are already part of many translation teams' workload. Yet these files often sit outside the document workflow. A PM uses one service for transcription, copies the result into Word, and then imports it into another system for translation. Along the way, client terminology, translation memory, and established review workflows are easily lost.

BabelBee's audio/video workbench brings transcription into the same workspace as document translation. Use it to transcribe media, edit timed text, label speakers, create a summary, and export TXT, DOCX, SRT, or VTT. If the client also needs a translation, send the cleaned captions to the translation workbench.

This article explains where the workbench fits, what it can produce, and when a transcript should move into translation.

Why transcription and translation should connect

Audio and video projects differ from ordinary documents. Meetings have multiple speakers, interviews need clear attribution, and subtitles must retain timecodes. The final deliverable may be a transcript, minutes, same-language captions, or translated subtitles.

When transcription and translation live in separate systems, PMs must coordinate multiple accounts, file formats, and versions. More importantly, a cleaned transcript cannot easily use the terminology, TM, and reviewers already available in the workspace.

The audio/video workbench closes that gap. First, turn spoken content into reliable timed text. Then decide whether to export it as-is or continue into translation.

Where to start

Open Audio-video workbench from the account menu, or choose Audio/video on New upload. After transcription finishes, the project appears in My projects like any other job.

Transcription uses the plan's workspace-minute allowance alongside its document-word quota. Duration is measured from the media file and rounded up to the next whole minute.

The workbench is designed for production transcription, caption review, and handoff to translation.

What you can do in the workbench

Upload an audio or video file. Transcription runs in the background against the workspace-minute quota. Duration is measured from the file; billing rounds up to whole minutes.

Edit speaker turns with a player. Cues are timed. Play a turn, rename a speaker, merge or split adjacent captions, and search and replace across the transcript. Automatic labels arrive as Speaker 1, Speaker 2, and so on. Similar voices, overlapping speech, and noisy recordings still need a human pass; when labels are unavailable, add names or roles during review.

Generate structured minutes from the edited transcript. Agenda, decisions, and actions remain separate from the cues, so the source transcript is never overwritten. Copy or download the result, and regenerate it after further edits.

Export TXT, DOCX, SRT, or VTT. A same-language transcript, a caption file, or a script for a reviewer can leave the workbench without starting a translation.

Send cleaned captions to translation when the deliverable is a translated transcript or localized subtitles. The translation workbench then applies the workspace glossary, TM, QA, reviewers, and either Standard or Agent production.

Automatic speaker identification is a starting point, not a substitute for review. Similar voices, overlapping speech, and noisy recordings may still require manual correction.

Meetings, interviews, and captions

Three common project types illustrate the workflow.

Meetings. A client kickoff, a vendor call, or an internal language-team review arrives as audio or video. The team needs a script they can search, speakers they can name, and sometimes minutes. Export TXT or DOCX for the record. Generate minutes when the client asked for decisions, not a full transcript. Translate only if the meeting must be delivered in another language.

Interviews. Research, journalism, and some HR work need named speakers and a clean script. Rename Speaker 1 and Speaker 2 before anyone quotes the file. Export DOCX for the researcher. Send to translation when the interview will be published or delivered in another language. Do not treat this workbench as a patient-records or certified-deposition service.

Captions. Media and e-learning clients still send files that must become SRT or VTT. Edit cues against the player, export the caption file, and send to translation when the subtitles themselves must be localized. The toolkit already converts subtitle formats as a standalone utility; the AV workbench is where you produce the timed transcript first.

In each case, ask one question: is the transcript itself the deliverable, or is it the source for a translation? Export handles the first; sending cleaned captions to translation handles the second.

When to send cues to the translation workbench

Not every transcript needs translation. A same-language record, internal minutes, or captions the client will publish in the original language can stop at export.

Send captions when:

• the client ordered a translated transcript
• the client ordered localized subtitles
• the same workspace must apply glossary and TM to the spoken text
• a reviewer must sign off on the translation, not only on the transcript

Clean the source first: correct obvious recognition errors, review segmentation, and rename speakers. The translation workbench processes the text it receives; it cannot determine whether an uncorrected name or recognition error is accurate.

Once cues are in the translation workbench, treat them as a document job. Choose Standard for straightforward captions or Agent when a long subtitle file needs stronger document-level consistency for names, terminology, figures, and repeated phrasing.

What to check before delivery

Automatic transcription can save substantial time, but client-ready work still needs review. Pay particular attention to names, terminology, numbers, dates, overlapping speech, and any passage that will be quoted publicly.

The utility toolkit includes subtitle-format conversion. That tool converts existing subtitle files; the audio/video workbench creates and edits timed transcripts from media. They serve different purposes.

How it fits the production workflow

A useful rule is to keep media and documents in the same workspace, while processing each in the workbench designed for it.

Monday: a client sends last quarter’s bilingual manuals (toolkit alignment + TM) and two kickoff recordings (AV workbench). The manuals go through Standard or Agent as usual. The recordings are transcribed, speakers renamed, minutes generated if needed. If the client also ordered translated minutes, send cues after the transcript is clean.

Mid-week: an interview must be translated. Edit the script in the AV workbench, export DOCX for the researcher’s files, then send captions so TM and glossary apply. Reviewers see the translation in the translation workbench, not in a pasted Word file.

Before delivery: caption jobs export SRT or VTT from the AV workbench. Translated captions follow the same export gates as any other Team Verified document job.

This preserves the player, timecodes, and speaker information needed for media while keeping terminology, TM, QA, and review available when translation begins.

Who this is for

The audio/video workbench is for LSPs and in-house language teams that already take meetings, interviews, or subtitle work alongside document translation, and want those files in the same workspace.

It is intended for teams that need an auditable path from media intake to reviewed transcript, captions, minutes, and translated delivery.

Frequently asked questions

Where do I open it?

From the account menu: Audio-video workbench. After a job finishes, open it from My projects.

How is transcription billed?

Against the workspace-minute allowance included with the plan. Plans include a monthly transcription allowance; see Pricing.

Can I export without translating?

Yes. TXT, DOCX, SRT, and VTT leave the AV workbench directly.

When should I send captions to translation?

When the deliverable is a translated transcript or localized subtitles, and after you have cleaned speakers and wording. The translation workbench then applies TM, glossary, QA, and reviewers.

Does Agent production run on the transcript itself?

No. Standard and Agent are document production methods, chosen on the segmentation-preview step of a translation job. Transcription is a prior, parallel step.

How good is speaker identification?

Labels are automatic (Speaker 1, 2, …). You can rename them. Overlapping speech will not be perfect. Plan a human pass on anything that will be quoted.

Can it replace post-production software?

The workbench focuses on transcript, speaker, caption, summary, and translation production. Use dedicated post-production software for picture editing, effects, and final media rendering.

---

Start a free workspace · Utility toolkit · Agent vs Standard production