How does everyone feel about local translator apps? I need a translator App for work with tts. I ditched google and got something called Rtranslator, and Sherpa tts which uses something called piper models.

I’m not sure what all of that amounts to, or if this any better than using google with llms? I can barely understand the language used when I try to learn what these components are. I’m not even sure if it is AI, or even the same kind of ai as something like google translate. Is is just as bad for the world as everything else? Or is it any better in any way other than being Foss and local? Do those things even matter when fighting ai?

  • Alex_Mihalchuk@piefed.social
    link
    fedilink
    English
    arrow-up
    5
    ·
    13 days ago

    Translation is exactly what AI can be used for. But I don’t need a machine translator telling me how to live my life and what to do.

  • solrize@lemmy.ml
    link
    fedilink
    arrow-up
    3
    ·
    13 days ago

    No idea if speech translation is done differently than a stt->translation pipeline. I sometime use Mozilla’s local translation built into current Firefox for text translation. It’s not as good as Google or DeepL at the moment, but I find it mostly usable.

    • ZDL@lazysoci.al
      link
      fedilink
      arrow-up
      3
      ·
      13 days ago

      Not as good as Google is pretty damning given just how shit Google’s translation is.

  • dumnezero@piefed.social
    link
    fedilink
    English
    arrow-up
    2
    ·
    12 days ago

    We should get to those eventually, it’s just not priority #1.

    Ask actual translators, humans. These translation models likely also suck in terms of getting training data WITH CONSENT.

  • TwilightKiddy@programming.dev
    link
    fedilink
    English
    arrow-up
    2
    ·
    edit-2
    12 days ago

    Neural networks have been used for language translation long before LLMs came to be. Natural languages are very hard to put into any “hard” structures, after all.

    The best we could do for a long time is train a neural network to do it. The interesting thing is, LLMs are quite good at translating stuff because the principle is more or less the same: find patterns in the original language, replace them with the same patterns in the target language. Just because of the sheer amount of data that goes into an LLM, it becomes good at translating stuff.

    That said, it’ll never be perfect. These things don’t understand any meaning behind any of the words, they just look for patterns. If you are fluent in more than one language, it becomes very apparent when you try to translate things back and forth. Often it will convey the same meaning, but it’ll sound a bit off. Sometimes it can completely hallucinate the context and then your translation will be way off.

    If it’s a professional setting where you can’t afford to get things wrong, do hire a professional translator.