Why the audio has to stay on your hardware
Qualitative research runs on consent, and consent comes with conditions: where recordings are stored, who can process them, and how quickly they are destroyed when a participant withdraws. Every cloud transcription or analysis service adds a processor to that chain — someone your ethics board has to vet, whose retention policy you inherit, and whose breach becomes your incident.
Kajo removes that step instead of documenting it. Transcription, speaker labelling, translation, search, and chat all run on the computer the recording is already on. Your protocol’s retention rules are enforced by you, on your disk, not by a vendor’s deletion queue.
Sample protocol language
Adapt this to your study; your ethics board or IRB decides what it needs.
Audio recordings of interviews will be transcribed and analysed using Kajo Voice, a desktop application that runs entirely on the researcher’s [institution-managed, encrypted] computer. Transcription, speaker identification, translation, and analysis are performed on that computer; no audio, transcript, or derived text is transmitted to the software vendor or to any other third party. Transcripts and analyses are stored in the application’s local database on that computer, protected by the institution’s device access controls and full-disk encryption; the application also keeps a compressed copy of each recording in its data folder on the same computer. The application makes network requests only to download its models (from the vendor’s server, or from Hugging Face when that server is unavailable), for software licensing and update checks, and to send feedback when the user chooses to. Its optional crash reports (sent to the vendor’s crash-reporting service, Sentry) and usage analytics are off unless the user agrees to them, and they remain switched off for this study. None of these requests carries audio, transcripts, or other research data. When a participant withdraws, their recording is deleted from the application, which removes that copy, its transcript, search index, and associated analyses; the original audio file is deleted separately in line with the study’s retention schedule.
Where Kajo fits in a research stack
| Stage | Tool | What happens there |
|---|---|---|
| Record | Your recorder or video-call tool | Kajo does not record. Capture with the tool your protocol approves, with consent. |
| Transcribe | Kajo | On-device transcription in 99 languages, speaker labels, and translation if the interview needs it. |
| Explore | Kajo | Search every transcript; ask the study folder for themes, quotes, and contradictions, with citations. |
| Code | NVivo, ATLAS.ti, or MAXQDA | Import the QDPX export: speakers arrive as codes, with exact timestamps. |
| Close the study | Kajo | Batch-export what you archive (Lifetime), then delete what your protocol says must go. |