How do I properly transcribe an interview?
Short answer
Clean recording + AI first pass + 30-60 minutes of editing per interview hour produces publication-ready transcripts in a fraction of the time.
A good interview transcript happens in three phases — prep, recording, and editing. AI now replaces the typing pass; it doesn't replace the care.
Preparation - One mic per speaker position: each person gets their own (lavalier or headset). This prevents reverb and speaker mix-ups. If you only have one mic, place it equidistant between speakers. - Choose a room with carpet, curtains, soft furniture — not a bare conference room. - Recording device: a dedicated recorder (Zoom H1n, H5) is more reliable than a laptop. Run a backup recording on your phone. - Set levels before you start — strong peaks without clipping.
During the interview - Open with a clean intro that names everyone ("Today I'm speaking with Maria Müller, …") — helps diarization later. - One person finishes before the next starts — avoid affirmative mm-hmms and yeah-overlaps. They wreck diarization. - On Zoom/Teams: capture local audio, not just the cloud version (which is compressed).
AI first pass - Upload to a provider like DeepScript, pick the Premium model for research interviews (diarization, higher accuracy), set the correct language, and add a custom vocabulary with proper nouns and jargon if relevant. - Processing: 2-5 minutes per hour of audio.
Editing AI transcripts are a first draft. Budget 30-60 minutes of editing per audio hour. Steps: 1. Verify speaker labels: AI sometimes swaps speakers on short interjections. Fix them and use real names instead of "Speaker 1" / "Speaker 2". 2. Cross-check proper nouns and jargon. 3. Choose a format: - Verbatim: every "um," every stutter, every pause noted. For conversation analysis, phonetics, court records. - Clean / smooth: filler words removed, grammar smoothed. For publication, book quotes, journalism. - Smart verbatim: middle ground — content complete but lightly cleaned. Default for qualitative research. 4. Insert timestamps every 1-2 minutes or at topic shifts. 5. Append a final paragraph with date, interviewer/interviewee, duration, location.
Export For research: DOCX with speaker labels and timestamps; for publication: plain text with structured paragraphs. DeepScript exports TXT, SRT, VTT, JSON directly — DOCX via Pandoc or Word.
Related questions
Still have a question?
Three transcriptions free to try. Or drop us a line — we answer within 24 hours, compliance questions included.