Better transcription AI than Otter?
June 2, 2023 1:58 PM Subscribe
I’ve got some audio that’s from a speaker with a strong accent and the sound quality isn’t great. Is there something better than Otter for getting a rough transcript that needs less correction? Thanks!
Jumping off of StrawberryPie's answer, if you're on a Mac, there's a program called MacWhisper that offers an easy frontend to the Whisper tech. I can't speak to whether Whisper is better than Otter for accents or bad sound quality, but I will say that Whisper, using the large language model, was able to transcribe a Hawaiian-language presentation for me with amazing fidelity, much more than Otter was able to handle.
posted by flod at 5:26 PM on June 2, 2023 [1 favorite]
posted by flod at 5:26 PM on June 2, 2023 [1 favorite]
speaking from experience, it might help if you can improve the audio quality. Depending on why it's bad (too quiet, too loud, hissy, popping, noisy, etc.) there may be things you can do to improve it before you feed it into Otter (or whatever transcription service you wind up using).
posted by sardonyx at 5:50 PM on June 2, 2023
posted by sardonyx at 5:50 PM on June 2, 2023
I recently used Sonix.ai for transcription from folks with many different accents and had great results.
posted by mdonley at 9:30 PM on June 2, 2023
posted by mdonley at 9:30 PM on June 2, 2023
Best answer: There's an online version of Whisper at Replicate that you can upload your sound files to. I just had it transcribe a 30 minute European Portuguese language course; took it 3 minutes. In my testing previously, Whisper did well with unusual accents, including a French / Australian English accent.
posted by Superilla at 11:57 PM on June 2, 2023 [1 favorite]
posted by Superilla at 11:57 PM on June 2, 2023 [1 favorite]
Response by poster: superzilla I could kiss you. This page is working much better than Otter for accented speech!
posted by The Last Sockpuppet at 10:52 AM on June 4, 2023
posted by The Last Sockpuppet at 10:52 AM on June 4, 2023
This thread is closed to new comments.
The following is not a terribly useful answer, but in case there are few other replies, here are 2 ideas. Microsoft offers speech recognition in Word on the web, and it looks easy to use, so it might be worth a try. OpenAI recently introduced Whisper, which is supposed to be state-of-the-art, but I'm having trouble finding an easily usable program or application that would let you use it. (But if you're comfortable with command lines and coding, the GitHub repository explains how to install it and run it.)
posted by StrawberryPie at 5:06 PM on June 2, 2023