Transcription service questions, answered plainly

Accuracy and quality

How accurate is it? Very good on clear recordings in English and other major languages. We do not publish an accuracy percentage because it depends on your audio. Noisy rooms, crosstalk and heavy accents lower it, which is why unclear passages are marked with a timestamp, like [inaudible 12:04], so you can check them.

Is a person listening? No. It is AI in two passes: speech recognition, then an editing pass for punctuation, paragraphs and speakers. That is why it costs a fraction of per-minute human transcription.

Does it guess when it cannot hear? No. Passages it cannot trust are flagged. Names it is told about (in your brief) are used in the spelling you give.

Verbatim or clean? Your choice. Clean read, the default, removes ums and false starts. Verbatim keeps everything.

Speakers and format

Does it identify speakers? It infers them from the conversation and the names you give it. On calls with many similar voices, check the labels. Tell us who is talking for the best result.

What do I get back? A Word document and a plain text file. SRT and VTT subtitles, a summary, action items and chapters when you ask. See how it works.

Can I get timestamps? Yes, per speaker turn by default or at a set interval such as every minute.

Can it make subtitles? Yes, SRT and VTT with up to 42 characters a line and two lines. See subtitles.

Files, languages and limits

What file types work? MP3, WAV, M4A, AAC, OGG, FLAC, MP4, MOV, WebM and MKV.

How big can a file be? Up to 4 GB each, and up to 10 files per job.

How long can a recording be? We do not set a minute limit; the stated limit is file size. Longer recordings take longer.

Which languages? English and most major languages, in the language spoken. It does not translate.

Can it do legal or medical transcripts? No. Use a certified reporter for court work. Please do not upload HIPAA-covered medical records.

Privacy

Is my recording private? Files pass through our upload server in the US, are stored and processed on our servers in Germany, and are read by our AI provider in the US under terms that exclude training. They are deleted after 30 days.

Price, turnaround and changes

How much does it cost? $25 a month for 20 files or $79 for 100. Details on pricing.

How fast is it? Usually well under an hour. A 4-minute test file took 80 seconds.

Can I ask for changes? Yes. Each job includes two rounds of changes.

What if it misses? You have 14 days to ask for a refund.

Can I cancel? Any time, from your account page. You keep access until the end of the paid month.

Can I try it? Start with the Starter plan and one real file. If the result is not what you asked for, ask for the refund.

Choosing between AI and human transcription

When is AI transcription enough? For interviews, podcasts, lectures, meetings and subtitles where you will read or skim the result, and where an occasional wrong word is fixable. It is the practical choice when you have many recordings.

When do I need a human? Court and regulatory transcripts, anything that must be certified, medical records, and recordings where one wrong word has real consequences. For those, use a certified service. Rev's own pricing page lists human transcription at $1.99 a minute, which gives you a sense of the cost difference.

Can I use both? Many people do: AI for volume, a human for the one recording that matters.

When something looks wrong

A name is spelled wrong all through the transcript. Reply with the correct spelling. Each job includes two rounds of changes, and the fix is applied everywhere.

Speakers are mixed up. Tell us who is who, and where possible add a few words one speaker says near the start. We relabel.

The transcript seems to stop early. The last timestamp is checked against the length of the recording before delivery. If you still see a gap, tell us the time it stops.

There are many [inaudible] marks. That is the audio, not a bug. Better microphones, less background noise and separate tracks per speaker help. If a job misses what you asked for, you have 14 days to request a refund.

The recording has two languages. We transcribe each language as spoken and do not translate. Mixed-language recordings are harder, so test with a short file first.

Using the document

Can I edit the Word file? Yes. It is a normal .docx with headings, speakers in bold and timestamps, and a plain .txt copy.

Can I get both a transcript and subtitles? Yes, from the same file. Ask in the brief.

How long do you keep my files? They are deleted after 30 days, so download your results.

More about Transcribe