Buy Transcription Services from Verified Specialists

The Complete Guide to Transcription, Subtitling and Captioning

Transcription looks like the simplest service on any marketplace and is quietly one of the easiest to buy badly. The deliverable is just text, so every provider looks interchangeable and price wins. Then the file arrives with no speaker labels, filler words stripped out of a research interview that needed them, and a scatter of guesses where the audio was hard.

Buying well is mostly specifying well. Six decisions shape the output — style, timestamps, speaker labels, file format, accuracy tier and turnaround. State those clearly and quotes become comparable.

What Human Transcription Actually Involves

A transcriber does not simply type what they hear. They segment continuous speech into sentences that never existed, place punctuation that changes meaning, research every proper noun, and mark honestly where the audio defeats them. The rule of thumb is three to five working hours per hour of clean single-speaker audio, so a much faster rate usually means lightly-corrected machine output.

Verbatim Versus Clean-Read

Full verbatim keeps everything, including hesitations and abandoned starts, because in legal proceedings, HR investigations and market research how something was said is itself evidence. Clean-read removes fillers while preserving every substantive point and the speaker's own vocabulary, which is the right default for podcasts and subtitles. What it must never do is paraphrase. State your choice in the order, since providers use these terms differently.

Timestamps, Speaker Labels and File Formats

Timestamps at fixed thirty-second or one-minute intervals are cheapest and enough for quoting and fact-checking; stamps at every speaker change suit interviews; word-level timing costs more and is for precise editing. Labels can be real names, roles such as Interviewer and Participant, or anonymised codes — if ethics approval requires anonymisation, say so before work starts so real names never enter the file. Agree the output format too, because a transcript destined to become subtitles is segmented differently.

Subtitles, Captions and the SRT or VTT Question

Subtitling is not a transcript with times attached. Each cue must fit on screen, stay up long enough to read and break at a sensible grammatical point; broadcast practice generally keeps a subtitle to two lines of around 37 to 42 characters. SRT is the universal interchange format, while WebVTT carries the positioning and styling that SRT cannot. Captions add speaker identification and meaningful non-speech sound, and burned-in captions cannot be restyled later, so agree font, size and safe areas per aspect ratio before rendering.

Why AI-Only Transcripts Still Fail

Speech recognition is excellent with one speaker, a decent microphone, a quiet room and a mainstream accent, and it makes a sensible first pass. Very little real audio looks like that. Accuracy falls away with regional accents and second-language speakers, with crosstalk a person can separate and a model turns into one garbled stream, and with the specialist vocabulary you most need right. The deeper problem is confidence: a human writes an inaudible tag with a timestamp, whereas a model outputs the most probable words, so an error looks exactly like a fact.

Accuracy Tiers and Turnaround by Audio Hour

A draft tier is machine output with a light human sweep, fine for search and internal notes. A standard tier is a full human pass, and it is the usual choice for anything you intend to publish or quote from. Some providers also offer a reviewed tier that adds a second pair of ears, and a few will supply a signed statement describing how the transcript was produced — whether such a statement satisfies a particular court, regulator or funder is a question for that body, not for the marketplace, so check before you commission. Treat any accuracy percentage promised against audio the provider has not heard as a warning sign. Turnaround is commonly quoted at a day or two for a one-hour file, faster at a premium, and large batches are usually quoted on total audio hours — agree it in the order rather than assuming it.

Legal, Medical and Academic Transcription

Specialist domains change the requirements, not just the vocabulary. Legal work means strict verbatim, careful speaker identification and frequent timestamps. Medical work means clinical terminology, drug names and dosages recorded exactly. Academic and qualitative research usually means anonymisation from the first draft and consistent conventions for pauses and overlaps so coding software can use the output. In all three, agree non-disclosure terms, storage and deletion before you send anything.

Accessibility, Compliance and How to Brief Well

Captions exist first for people who are deaf or hard of hearing, and accessibility guidelines treat them as a baseline requirement on pre-recorded video rather than an enhancement. The side benefits are real: much social video is watched with the sound off, and a published transcript gives search engines something to index. A good brief is short and specific — duration, speakers, recording conditions stated honestly, style, timestamps, labels, formats, deadline and confidentiality — with a glossary attached and the hardest minute of audio shared rather than the cleanest.

Ready to start? Use Zinn Finder, or explore related services such as translation services, proofreading services, voice over services, content writing, UGC creators and Filipino video editors.