Using computer vision to read the existing tab from the page would be more efficient and precise than trying to use audio processing to isolate individual notes from a single instrument in a potentially busy audio track. The existing tabs also nail down which string and fret is used for each note, which would be a difficult task for audio processing because there are multiple combinations of string and fret for any given note. For example, an open G can also be played on the 5th fret of the D string, or the 10th fret of the A string, or the 15th fret of the E string.

I understand but would assume it could be solved by chord/positional groupings or similar. Manual/human tab transcription need to infer positional assumptions anyway; some artists will play the same songs different ways too ala Bob Weir

> the existing tab from the page would be more efficient and precise

But what about songs that aren't made into a YT video?