Transcription Software for Interviews and Qualitative Research
Data checked:
For research interviews the first question is not accuracy but where the audio goes. If your study has an ethics approval or a consent form, the tool has to match what you told participants about where recordings are processed and stored, and whether they can be used to train AI. After that, check speaker labels, how long a single recording can be, and whether the export fits your analysis software.
Option 1: keep the audio on your computer
On-device apps never upload recordings for transcription. All three below label speakers.
| Tool | Platforms | Paid from | Speaker labels | Export | Languages |
|---|---|---|---|---|---|
| MacWhisper File transcription, Offline / on-device | Mac | €64 one-time (Pro) | Yes | TXT, SRT, VTT, DOCX, PDF, MD, HTML, CSV | ENJAZH 100 languages |
| Vibe File transcription, Offline / on-device | Mac, Windows, Linux | Free | Yes | SRT, VTT, TXT, HTML, PDF, JSON, DOCX | ENJAZH |
| Buzz File transcription, Offline / on-device | Mac, Windows, Linux | Free | Yes | TXT, SRT, VTT | ENJAZH 100 languages |
See offline transcription software for the differences between them.
Option 2: cloud services that don’t train on your files
| Tool | Paid from | Speaker labels | Export | Max length per file | Trains on your data? |
|---|---|---|---|---|---|
| Sonix File transcription | $25/mo $22.92/mo billed yearly (Core) | Yes | DOCX, PDF, TXT, SRT, VTT | Paid: Not verified | No |
| Happy Scribe File transcription, Meeting notes | £17/mo £8.50/mo billed yearly (Basic) | Yes | TXT, SRT, PDF, DOCX, VTT, HTML | Paid: Not verified | Only if you opt in |
| tl;dv Meeting notes | €29/mo €18/mo billed yearly (Pro) | Yes | Not verified | Free: 3 h Paid: 3 h | No |
- Sonix says it never uses your audio, video or transcripts to train its models, and deleting a file wipes both. It sells pay-as-you-go hours ($10 an hour), which suits a fixed batch of interviews.
- Happy Scribe uses content for training only if you opt in at signup. Its Basic plan caps recordings at 90 minutes; Pro and Business have no per-recording limit.
- tl;dv is built around recording video calls; free users can upload up to 5 files.
Option 3: popular services that may train on your recordings
| Tool | Paid from | Speaker labels | Max length per file | Trains on your data? |
|---|---|---|---|---|
| Otter.ai Meeting notes, File transcription, Live transcription | $16.99/mo $8.33/mo billed yearly (Pro) | Yes | Free: 30 min Paid: 4 h | Yes |
| Rev File transcription, Meeting notes | $29.99/mo $25.49/mo billed yearly (Essentials) | Yes | Free: Not verified Paid: Not verified | Yes |
| Descript File transcription | $24/mo $16/mo billed yearly (Hobbyist) | Yes | Free: Not verified Paid: Not verified | Yes, unless you opt out |
| Fathom Meeting notes | $20/mo $16/mo billed yearly (Premium) | Not verified | Free: Not verified Paid: Not verified | Yes, unless you opt out |
Otter.ai trains on de-identified audio and transcripts. Rev’s terms say customer content is used for continuous training of its speech models. Descript and Fathom train on your data unless you opt out. If you use one of these, check that your consent form allows it and switch off sharing where you can.
Exporting to NVivo, MAXQDA or ATLAS.ti
None of the vendor pages we checked documents a direct export into NVivo, MAXQDA or ATLAS.ti. The usual route is a DOCX or TXT transcript with speaker names and timestamps, which those programs import. Notta describes its TXT, DOCX, XLSX, PDF, SRT and VTT exports as suitable for these tools.
Languages
For Japanese or Chinese interviews, check each tool’s language list in the directory: Descript does not transcribe either, and Fathom supports Japanese but not Chinese.