mrkn1

109 languages, including other alphabets.

Just ran

  textsnap "https://i.ytimg.com/vi/LBNDfxjEYlA/maxresdefault.jpg"

and got this

  $('.count').each(function () {
  $('this').prop('Counter', 0).animate({
    Counter: $('this').text()
  }, {
      duration: 4000,
      easing: 'swing',
      step: 'function (now) {
          $('this").text(Math.ceil(now));
      }
    }); 
  });

thank you!

thank you! what is it about?

I have not. My experience has been a few seconds for an 1024x1024 with medium density of text, FWIW. Feel free to try it on a few test images, model is pretty small and fast, but yeah no formal evals on CPU.

It should support 109 languages. More info here: https://huggingface.co/PaddlePaddle/PaddleOCR-VL

No, simply, my laptop only has a CPU.

I haven't. But the evals of the underlying model are published here, including on Omnibench. https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5

For 100% local CPU fact checking, I made this: https://news.ycombinator.com/item?id=48301003

A lot of my queries are summarize/explain/fact check, and these are covered 100% on my CPU locally [0], reducing frontier model reliance

[0] https://news.ycombinator.com/item?id=48301003

just released new version that implements your ideas

new version released is up to 3x faster on CPU. Let me know!

thank you being thorough

clipboard: rn input is treated like any other source, so text gets written to ./textsnaps/clipboard_ocr.txt, and stdout just prints that path. Nothing goes back to the clipboard in this version (stay tuned)

portability: agreed, and it's a small change. textsnap already looks for the checksum manifest next to the script before falling back to the cache, so extending it should be easy. I make a note for next version.

Great question. I'm not familiar with docling-serv but pretty different beasts from what I gathered. Docling is a heavier pipeline (actually uses GPU).textsnap is the opposite: single-file CLI, small VLM running on plain CPU cores, one command, no server. Tradeoff is CPU decode is sequential so it's slower on dense pages, and it OCRs one image rather than doing full layout.

If docling-serve is already meeting your needs it's probably not an upgrade. But it installs in one command, so would love to hear how it stacks up on your images, if you end up trying it.

thanks! yapsnap is audio to text, and textsnap is image to text. Both have been daily use cases for me for a while. And yes, the feedback on yapsnap encouraged me to also release textsnap on github

Reading this means a lot, thank you! Even faster version in the coming days, stay tuned and PRs welcome!

Added Diarization / Speaker Separation that is fast and CPU only. Thank you all for the great feedback and support. PRs welcome!

  yapsnap "https://www.youtube.com/watch?v=NzKJ-xO-VhE" --diarize

  SPEAKER_00 [00:00]: Welcome to the show.
  SPEAKER_01 [00:03]: Glad to be here, thanks for having me.
  SPEAKER_00 [00:08]: Let's get started.

done! new version separates speakers on CPU fast

done! just pushed a new version with CPU diarization

Kroko's website says benchmarks aren't formalized yet. FWIW, this url says 5% WER for English [0]. though it doesn't specify the dataset, so not directly comparable to Parakeet's 6.32 on the Open ASR Leaderboard

Best way to judge is to try it on your own audio

[0] https://huggingface.co/hudaiapa88/sherpa-stt-onnx

in the roadmap!

will work on it, that would be neat. I love pyannote but not happening on CPU at reasonable speeds lol

thank you! good use case, what hetzner box specs have you chosen?

small, ONNX-optimized models designed specifically for low-latency CPU streaming, so it avoids overhead of large transformer arch and GPU memory transfers

thanks! making a note of the feature request

Good point! I haven't found a faster way to consume info than reading. But depends on the type of learner you are (visual, auditory, hands-on/interactive, etc)

thank you!

Just download the model for your preferred language, all hosted on the Kroko-ASR collection here: https://huggingface.co/Banafo/Kroko-ASR/tree/main Right now you have Dutch, French, Portuguese, Spanish, German, Italian, Swedish, Swiss German, Hebrew, and Turkish. Grab the one that matches your audio, point yapsnap at it with --model (or set KROKO_MODEL), and you're set!

Yes fair point, asr cached and exposed. I meant to draw the line more on fetchable or not.