Journalists, researchers and recruiters all hit the same wall: the interview went well, and now there is an hour of audio that has to become text before anything can be written. Done by hand, that is most of a day. Done properly, it is under an hour.
Most of the saving comes from decisions made before the interview starts.
Before: record so the transcript is easy
- One device, close to both people. A phone on the table between you, not in a pocket. If you are on a call, record through the platform rather than pointing a microphone at a speaker.
- Kill the room noise you can. Close the window, move away from the coffee machine, ask for the fan to be turned off. Steady noise is survivable; sudden noise is not.
- Say the names at the start. "This is a conversation with Dr. Anna Kowalski about..." — it takes four seconds and gives you the spelling anchor for the rest of the file.
- Do not talk over each other. The hardest thing to ask and the most valuable. Overlapping speech is where every transcription method fails, human or machine.
- Get consent on the recording itself. Recording rules vary by country, and having the yes captured in the file solves the argument before it happens.
During: mark the moments that matter
When the interviewee says something you know you will quote, note the time. A scrap of paper with "14:20 funding" is enough. In editing you jump straight there rather than scrubbing through an hour looking for a sentence you half-remember.
After: the transcription pass
Upload the file, choose the language, and let it process. A one-hour interview typically takes a few minutes. If both people speak the same language throughout, leave detection automatic; if the conversation switches languages, set the output language explicitly.
For interviews specifically, speaker identification is worth having — it is a Pro feature on SpeakToWords — because a two-person transcript without labels has to be re-attributed by hand, and that is slower than it sounds.
After: the editing pass
Work in this order. It is faster than reading top to bottom:
- Names and terms first. Find every proper noun and fix the spelling once. Machine errors are consistent, so one replacement usually corrects every instance.
- Then your marked timestamps. Check the passages you intend to quote against the audio, word for word. These are the only places where an error actually costs you.
- Then structure. Insert a heading every time the subject changes. A 9,000-word transcript with twelve headings is navigable; the same text without them is not.
- Filler last, and only if publishing. If the transcript is for your own use, leave it. Cleaning speech is cosmetic work.
Quoting accurately
Two rules keep you out of trouble. First, quote from the audio, not from the transcript — always listen to the passage you are putting in quotation marks. Second, if you tidy a quote for readability, tidy only the filler; changing word order or vocabulary changes meaning, and a recording exists that says so.
Archiving
Keep the audio as long as you might need to defend a quote. Keep the transcript somewhere searchable — six months later, "which interview did she mention the 2019 merger in?" is a question you can only answer with full-text search.
The free tier covers 15 minutes a month, which is enough to transcribe a short interview and see whether the workflow fits how you work. For a steady flow of hour-long interviews, Basic at 500 minutes per month is the realistic plan.