Why researchers end up here
Three things collide in qualitative research: an ethics approval that promises participants confidentiality, a stack of recordings that need to become text, and a budget. Transcription services solve the second at the expense of the first and the third: the audio goes to a company, often abroad, at a price per minute. Doing it yourself by hand solves the first and third and costs weeks.
Local transcription is the fourth option. The model runs on your laptop, the recording never leaves it, and the cost is flat. What you give up is the polish of a human transcript, which matters for some methods and not for others.
A local research workflow
- Record well. A recorder or phone between you and the participant. Keep the original file; do not re-export or compress it.
- Transcribe locally. Drop the file into Scrieb. Speaker identification separates interviewer and participant, and handles focus groups of up to about six people.
- Check the labels. Fix the few places where speakers overlapped. Far faster than labelling from scratch.
- Anonymise on your machine. Replace names and identifying details before the transcript leaves your computer for the analysis software or your co-authors.
- Export and import. DOCX or TXT into NVivo, MAXQDA, ATLAS.ti, Dedoose or a spreadsheet.
What the ethics board will ask
Where is the audio processed?
On the researcher’s own computer. No transmission to any service, so no third-party access and no cross-border transfer.
Who can access the transcripts?
Only whoever can open the machine. Store the project folder on the encrypted, backed-up volume your data management plan specifies.
How is data deleted at the end of the study?
Delete the recordings in Scrieb, which removes transcripts and notes, and optionally the audio files. Standard deletion; use your institution’s secure-erase policy for the disk if required.
Is any content used to train AI?
No. The models are downloaded, not trained by us, and no content reaches us at all. Details on the security page.
Software or service: a quick comparison
| Local software (Scrieb) | AI transcription service | Human transcription service | |
|---|---|---|---|
| Where the audio goes | Stays on your computer | Uploaded to the vendor | Uploaded, then heard by a transcriber |
| Cost for 25 one-hour interviews | One subscription | Typically $150–$400 | Typically $1,500–$3,500 |
| Turnaround | Minutes to hours, unattended | Minutes to hours | Days |
| Speaker labels | ✓ | Usually | ✓ |
| Verbatim with disfluencies | Manual for key passages | Rarely | On request, at a premium |
| Consent form fit | No third party | Third party, often abroad | Third party plus individual transcriber |
Full cost breakdown: Transcription services vs software.
Pricing
Flat price with no minute limits, so 5 interviews and 50 cost the same. Free trial first.
Questions from researchers
- Can I transcribe research interviews without uploading them?
- Yes. Scrieb runs the speech model and speaker identification on your own computer. The audio file is read from your disk and the transcript is written back to it. No server is involved, which is what most consent forms and ethics approvals assume.
- Does it work with NVivo, MAXQDA or ATLAS.ti?
- Export as DOCX or TXT and import into any of them. Speaker labels are preserved, so a transcript imports as a dialogue and can be auto-coded by speaker where the software supports it.
- How accurate is it for research interviews?
- With clear audio and a good microphone, Whisper-based transcription is comparable to cloud services, which mostly run the same models. Accented speech, jargon and poor recordings reduce accuracy for every tool. Plan a read-through with the audio for the passages you analyse closely.
- Is the transcript verbatim?
- By default Whisper produces clean text, without fillers and false starts. For thematic analysis that is usually what you want. For conversation or discourse analysis you will need to add disfluencies by hand for the relevant passages; no automatic tool does that reliably.
- How long does a one-hour interview take?
- A few minutes on a computer with a modern GPU, roughly an hour on a laptop without one. Either way it runs unattended and offline.
- What does it cost for a dissertation with 25 interviews?
- The same as one interview: Scrieb is a flat subscription with no minute limits. A transcription service at typical per-minute rates would charge several hundred to a few thousand dollars for the same set.