OpenAI's Whisper

tested it on the final battle scene from 8 Mile
is it good or shit?

ChatGPT Wizard Shirt $21.68

Beware Cat Shirt $21.68

ChatGPT Wizard Shirt $21.68

  1. 1 year ago
    Anonymous

    Pretty cool.

  2. 1 year ago
    Anonymous

    Its understanding of grammar is impressive, I haven’t seen any other transcriber on that level.
    Funny how its most noticeable mistake is a pretty easy one (ain’t no such thing as halfway crooks), but it nails the last part where Eminem is speaking quickly without music. Probably because of the crowd’s noise.

  3. 1 year ago
    Anonymous

    where does it store the downloaded models? Linux

    • 1 year ago
      Anonymous

      oh it's ~/.cache/whisper

  4. 1 year ago
    Anonymous

    Is there some place I can download the models and put on a flash drive? My internet is not good at home.

    • 1 year ago
      Anonymous

      You can create a docker image, save it (docker save), and load it (docker load) whenever you want.

  5. 1 year ago
    Anonymous

    it's not "pops" though. it's pac.

  6. 1 year ago
    Anonymous

    >pops
    pac.
    >but all six of you jumped
    by all six of you chumps
    >we did frick my girl
    wink did frick my girl
    >this hot plate groups
    as halfway crooks.
    >frick a pop, a doc
    frick papa doc

  7. 1 year ago
    Anonymous

    Model where

  8. 1 year ago
    Anonymous

    parents doesn't rhyme with marriage

  9. 1 year ago
    Anonymous

    Is this another of these shits I can't run if I don't have dedicated gaymer GPU?

    • 1 year ago
      Anonymous

      You can run with CPU but it's slower

      • 1 year ago
        Anonymous

        THANK GOD, god bless you anon

  10. 1 year ago
    Anonymous

    How difficult would it be to write a program to use this for speech to text? I'm a victim of US schools and never learned how to spell. To this day I sometimes need to use Google voice typing for some words and have been looking a first programming project.

    • 1 year ago
      Anonymous

      >and have been looking a first programming project
      >looking a first
      Less effort to learn to spell (and some grammar apparently).
      Erm, can I translate, let me try:
      You finna speak first fool, then spell `my homie`.

      • 1 year ago
        Anonymous

        you're a loser for having the time to respond like this

      • 1 year ago
        Anonymous

        >Less effort to learn to spell
        I have been told I my entire life I can't. I have been looking for a way to improve my spelling for a long time and the best I can do is fly-spell in Emacs. I will still misspell some words so badly I end up using my (non-free) phone to spell it for me.
        In the past, I was dependent on Dragon Naturally Specking but now I only need my phone or search engine about every other sentence. Most normal gays think this is an acceptable way to live.

  11. 1 year ago
    Anonymous

    Very excited for this. I have some archived videos and always forget where I heard something from. With this can probably create a transcipt for each video next to it, and then index over all of them to create something searchable.

  12. 1 year ago
    Anonymous

    It's pretty good from what I've tested

  13. 1 year ago
    Anonymous

    bretty neat, could be really cool to have autogenerated subtitles on every video/lyrics on all music/etc. because of a simple addition/plugin to various websites and programs.

  14. 1 year ago
    Anonymous

    there is gpu cuda with is way faster. Too bad it can only handle up to medium on my 3070
    Does Japanese ok too

  15. 1 year ago
    Anonymous

    OP should test Whisper on those fast speaking commercials you see on TV or hear on radio. What about livestock auctioneers?

    • 1 year ago
      Anonymous

      I tried it with this:

  16. 1 year ago
    Anonymous

    .. does it work on porn?

    • 1 year ago
      Anonymous

      anything that has people speaking, yeah

      • 1 year ago
        Anonymous

        i have several terabytes of porn to run this on now

        4090 it is i think

  17. 1 year ago
    Anonymous

    Don't work on riscv

  18. 1 year ago
    Anonymous

    how can we know if it wasn't trained on that specific song

  19. 1 year ago
    Anonymous

    Could it be used to autogenerate literal subtitles for anime?
    >generate japanese subtitles
    >substitute each japanese word with the closest english equivalent or an AI generated glyph

    • 1 year ago
      Anonymous

      yes, "--task translate" gives you a subtitles file

    • 1 year ago
      Anonymous

      Yes it already can do that. It doesn't work like that though.

Leave a Reply to Anonymous Cancel reply

Your email address will not be published. Required fields are marked *