Can you point me to a safe automated transcipton method?

March 25, 2024 9:29 AM

I'm a writer and I often conduct interviews and then want them transcribed. These are just for me, so I can pull quotes from them, etc, so the transcriptions don't have to be neat or perfect. But now I'm concerned about privacy, and want to know the best options for preserving confidentiality.

I used to frequently use Rev, but then one of the places I write for told me to stop using it because apparently they overlap with Chat GPT, and thus confidentiality was at risk. I need to avoid this problem and preserve the privacy of these interviews. Do you know the best way for me to do this? Free or very cheap recommendations, please!

posted by gigondas to Technology (9 answers total)

Some of my library colleagues have recommendations in their Oral History Toolkit. At least a couple of those services highlight privacy.
posted by pantarei70 at 10:09 AM on March 25 [2 favorites]

Rogue Amoeba's Audio Hijack has a transcribe block. Mac, and not free. Excellent software.
posted by terrapin at 10:11 AM on March 25

Whisper is one we've been using a lot too -- Mac native and the basic license is free.
posted by pantarei70 at 10:12 AM on March 25 [3 favorites]

I use Otter.ai for this. It's free for individuals. Here is some information on their privacy policy.
posted by guessthis at 10:24 AM on March 25 [1 favorite]

Came here to also recommend Whisper, it's FREE and it can run COMPLETELY offline. You need sufficient hardware (a new Mac or a Windows machine with an Nvidia GPU) but you can use pretty much any low end Nvidia GPU so even if you were building a machine from scratch you could do the whole thing for under $500.

Whisper isn't the friendliest program (it's a command line program), but there is a front end for it (for Windows) that makes it a little easier to use. https://github.com/Purfview/whisper-standalone-win .

There are also detailed instruction on how to make Whisper run fully offline (no Internet connection at all, after the initial download and install) here
https://github.com/nicholasgcotton/WhisperDO/blob/main/Whisper%20Install%20(Offline).md (normally Whisper uses the internet to download the language model and some other data regarding word usage, but you can force it to do that ahead of time and then use it in a way you know 100% it's not calling home).
posted by tiamat at 10:38 AM on March 25 [1 favorite]

I also use Whisper and I wrote up how and why, especially regarding privacy and other ethics concerns.
posted by brainwane at 11:19 AM on March 25 [8 favorites]

I don't think you even need the GPU for Whisper, it will run in CPU-only mode (albeit slower and using a ton of system RAM for the AI model instead of GPU memory).

In my experience it does pretty well overall, although for my use (subtitling movies) it frequently ignores dialogue and mistimes stuff. Probably you won't have that issue with interviews.
posted by neckro23 at 11:32 AM on March 25

Response by poster: Thanks so much, everyone. Whisper for the win.
posted by gigondas at 12:29 PM on March 25

Ah, it looks like Otter has adjusted their free quota up from when I was using it to transcribe the Mefi podcast. I started using Whisper instead. If you're on a Mac, check out MacWhisper for a friendlier interface. (I found this, of course, after working out a command line workflow.)
posted by Pronoiac at 7:10 PM on March 25