Transcribing testimonies and interviews

A one-hour interview takes four to six hours to type out by hand. That is why transcription is the task where AI saves the most obvious time, and why documentation teams reach for it first.

The tools differ enormously by language, and by where your audio goes. Arabic transcription is usable. Kurdish Sorani is not: Whisper, the most widely used transcription model, does not list Kurdish among the languages it was trained on at all, so what it produces from Sorani audio should not be trusted even as a rough first draft.

Watch out for this

A recording of a witness is usually the most sensitive file your organisation holds. Uploading it to a website hands a copy to a company you have no agreement with, in a country whose courts you cannot reach. Whisper on your own computer avoids that: it downloads once and then works with no internet connection, so the audio never leaves the machine. Install it on the computer that stays in the office, and never upload raw testimony to a free online site.