20
Anyone using "Speech Note" (speech to text) with good results? (lemmy.ca)
submitted 1 year ago by to c/linux@lemmy.ml
5 comments fedilink hide all child comments
 

I've been using Speech Note (github link) for months, but it often gets things wildly wrong.

I thought it was my mic, so I got one that's crystal clear. I also tried a ton of different models, and other than being slow (or fast), their accuracy is usually pretty similar.

But I'm still needing to take a lot of time to edit the results, and I wonder if there's something I should be doing to get better results.

On other speech-to-text platforms (like Futo keyboard on Android), the results are fast and very accurate. I have a hard time believing that Speech Note can't be as good.

Can any other users share their experience?

UPDATE: Ok, the best model that I've found for Speech Note is the WhisterCpp FUTO English-244, which, funny enough, is the model I use on Futo Keyboard for Android. It's not the fastest, but fast enough. It is quite accurate, and that means less time editing text.

all 5 comments

sorted by: hot top controversial new old
[–] 4 points 1 year ago

Had enough issues with it to not find it helpful. But I'm not a native English speaker and support for my local language is so-so, so might as well be me that's the problem.

  • permalink
  • fedilink
  • source
    •
    [–] 3 points 1 year ago

    I haven't used Speech Note, but I have been using Whisper with great success. I run it via Docker.

  • permalink
  • fedilink
  • source
    •
    [–] 1 point 1 year ago (1 child)

    I've used it for a short while to test it out. Accuracy was pretty good, as was correct punctuation. Response time also good.

    It's using my Nvidia GPU to do the LLM thing, so that may be the difference.

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
  • [–] [S] 2 points 1 year ago

    It’s using my Nvidia GPU to do the LLM thing, so that may be the difference.

    This could be!

    Interestingly enough, I was playing around with LLama, as they have speech to text to interact with their chat bot, and it converts in near real-time with very good accuracy. So I do know that things can be fast and accurate, but I wish it was in Speech Note. LOL

    For now, I may just to STT through my phone on a shared document with my laptop.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 2 points 1 year ago

    Try a few different accents out - but I've never had better than a 95% success rate myself

  • permalink
  • fedilink
  • source
    •