Live data from Hacker News

Show HN: YouTube Guitar Tab Parser

github.com

61–70 of 72 posts

Re: Show HN: YouTube Guitar Tab Parser

#62
post #35

So I'm a bass guitarist, and I've made some of these exact videos. I link to a Patreon in my videos that lets you pay a membership to download the full tabs in PDFs. Here I was thinking AI might soon replace my painstakingly slow tab transcription efforts altogether. But I never thought about someone just ripping it from the video... Let me know if it works ;-) https://www.youtube.com/watch?v=c2i-Lwoe2Ow

Doesn't work perfect yet [1], probably needs some finetuning to get the bounding boxes correctly. Sorry for ripping it!

1: https://filebin.net/ylchm28lpbciqc5u

Re: Show HN: YouTube Guitar Tab Parser

#63
Claude vision seems like a massively overpriced VLM choice given Gemini has great VLM tower since 2.4 and my gut feeling is this is a task that local Qwen’s would handle easily.

Disclaimer: my daily workflows include millions of Claude and Gemini tokens spent on dev and video vision intelligence.

Re: Show HN: YouTube Guitar Tab Parser

#64
post #25

Earlier quoted context omitted.

I don't have an answer, but wanted to say that I started using Soundslice about 9 months ago and I absolutely love it. It does exactly the part of transcribing that I want done for me and nothing more, which is basically none of it except the boring stuff.

Care to sell me on it? How dose it compare to musescore?

Soundslice lets you start with a real recording (MP3 or YouTube) and integrates the audio/video playback directly with a tab/notation editor.

So as you play the audio (with slowdown, looping, etc.), you can immediately write down what you figured out, and that written notation stays synced with the original recording. This workflow for transcription is fantastic, because otherwise you're juggling multiple apps (audio playback/slowdown, plus notation software).

Since you asked about MuseScore: MuseScore is purely scoring software, without the ability to sync with real audio. You can also use Soundslice as a pure sheet music editor; the main difference is it's web-based as opposed to being a desktop app like MuseScore. Soundslice also has the philosophy that the music might always be displayed on different devices, so it's engraved on the fly depending on your screen dimensions. (MuseScore sort of assumes the end result will be a fixed-dimension such as paper.)

Re: Show HN: YouTube Guitar Tab Parser

#65
post #47

I learned this one: https://m.youtube.com/watch?v=07N4VVSmYLU&list=RD07N4VVSmYLU... …exactly the way you described. Pause, screenshot, adjust color/contrast, assemble in a pdf. Super annoying.

Learning music from recordings (video or just audio) would be so much easy if modern video and audio playback software would bring back the AB repeat function that used to be common on CD players, DVD players, LaserDisc players, and MP3 players. Why the heck did this largely go away?

Check out Soundslice for exactly this. It's great audio/video playback (with slowdown, looping, etc.), with an integrated tab and sheet music editor. So you can write down what you've figured out, and it remains synced with the original audio.

Re: Show HN: YouTube Guitar Tab Parser

#66
post #63

Claude vision seems like a massively overpriced VLM choice given Gemini has great VLM tower since 2.4 and my gut feeling is this is a task that local Qwen’s would handle easily. Disclaimer: my daily workflows include millions of Claude and Gemini tokens spent on dev and video vision intelligence.

This was what I had available, but adding support for other models/providers should be trivial. Adding support for local models is probably a bit harder and I don't have any experience with it. I doubt anything good enough will run on my macbook m1 though.
Post reply on HN