How to transcribe research interviews
Updated 2026-07-28
Transcription is where qualitative research wins or loses hours. A one-hour interview typed by hand costs four to six hours; with automatic transcription the work shrinks to reviewing and annotating, but only if the recording and the conventions were planned in advance.
This guide covers what to decide before recording, during transcription, and before you start coding.
Recording with transcription in mind
Put the microphone close to the speaker and away from fans, air conditioning and reverberant surfaces. Constant background noise degrades recognition more than low volume does.
In focus groups, ask participants to identify themselves the first few times they speak, and discourage cross-talk: no system separates two simultaneous voices well. Record in MP3 or M4A and check the file before the session ends.
Verbatim, natural or reduced
Verbatim transcription keeps repetitions, fillers and pauses, and is essential in discourse or conversation analysis. Natural transcription cleans hesitations without altering content, and is the usual choice for thematic analysis.
Reduced transcription keeps only what is relevant to the research question. It saves time, but it is an analytical decision: document the criterion in your methods section, because it constrains what you can later claim.
Minimum conventions
Decide before you start how you will mark speaker turns, long pauses, overlap, inaudible stretches and researcher comments. A simple scheme applied consistently beats an exhaustive one applied halfway.
Add timestamps every few minutes: when a quotation turns out to be central, you will want to return to the audio to check tone and context.
Anonymisation and data protection
Replace names, places and institutions with stable codes from the first version of the text, and keep the mapping table separate from the corpus with restricted access.
Keep the informed consent, state in it that recording and transcription will take place, and apply the retention policy you declared to your ethics committee. In European projects, check the legal basis for processing before uploading audio to any service.
From text to analysis
Once the transcript is reviewed, read the whole set before coding: the first complete reading usually reshuffles the planned category system.
Work with plain, exportable text. If you later move the corpus into another qualitative analysis tool, you will be glad the material is not trapped in a proprietary format.
Frequently asked questions
- How long does reviewing an automatic transcript take?
- Between a third and a half of the audio duration when the recording is good. With background noise or overlapping voices it can approach the cost of typing it yourself.
- Can I transcribe in several languages?
- Yes, although switching language within a sentence is the hardest case. If your corpus alternates languages, review those passages with particular care.
- Should I transcribe every interview in full?
- Not necessarily. With large samples it is common to transcribe part of the corpus in full and the rest in reduced form, always declaring the criterion in the methods.