Computer-generated transcription that won't steal data?

February 10, 2025 3:32 PM

I'm working on an ethnography project with a nonprofit organization that involves interviewing community members about potentially sensitive information. I'd like to use a free AI-transcription service to do the bulk of the transcription so I can focus on making corrections and other aspects of the work. Looking at services available, I'm guessing that most would save my recordings/transcripts for their own data mining purposes, which I don't want. Is there a free or low-cost (<$100 USD) one time payment service that will guarantee not to keep my files?
posted by Summers to Technology (3 answers total)

You could try using any number of tools that run locally on your device instead of sending your audio somewhere else. MacWhisper, for one.
posted by emelenjr at 3:40 PM on February 10 [5 favorites]

The OpenAi API interface doesn't use your data.

Though I think emelenjr's suggestion for using local Whisper is really good.
posted by gregr at 4:22 PM on February 10

i purchased macwhisper recently for a project where, likewise, i did not wish to feed the ai monsters with my good datas. everything is local to my machine (M2 MBP). accuracy with the pro models is quite good; free version accuracy is ok/fine. be aware you will need to manually code voices if there is more than one speaker on your audio track; this is not an automated function. cost exchange to my currency was still well under 100USD. app has been impressive in many ways, but their help docs are very slim.

before i purchased macwhisper i was also looking at aiko, which is free. i did not actually use it, just mentioning as something that may be worth investigation.

if you are not running a mac, well, hopefully someone else can chime in. a colleague suggested google cloud speech to text had some opt-out options but i did not investigate far enough to suggest. godspeed to you!
posted by tamarack at 9:13 PM on February 10