In qualitative research, transcription is primary data. Transcription errors become analysis errors. Accuracy verification is not optional — it's part of the methodological protocol.
Sources of Error in Automatic Transcription
- Proper names: uncommon people, institutions, places
- Technical or disciplinary terminology: specialized professional jargon
- Strong accents or regional dialects
- Poor audio quality: noisy environments or poor microphones
- Speaker overlap: when two people speak simultaneously
Recommended Verification Protocol
20% stratified random sample
For projects with 20+ interviews, verify 20% via stratified random sample. If average accuracy exceeds 95%, the full corpus is acceptable with a methodological note.
Full verification of critical interviews
Interviews with the most relevant participants or most important theoretical contributions must be verified 100%, regardless of overall accuracy.
Process documentation
Record the tool used, model version, average error rate of your sample, and correction process. This information goes in the methodology section.
For ethics committees and reviewers: most academic journals now accept verified automatic transcriptions. Document the process with the same rigor as any other methodological decision.