-
Hello |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 2 replies
-
Ah, ah! OK, perhaps it's time to give tablature recognition a second chance. |
Beta Was this translation helpful? Give feedback.
-
I was used to Marcel Dadi's books of good quality with fret + stems + beams. The current Audiveris version uses various methods to process a partition, depending on the items to handle. For instance, note heads use template matching, beams use image closing, texts use an external OCR software, stand alone symbols use a neural network trained on representative glyphs, etc. Regarding tablatures, I think OCR would not be the key component, because it is meant to detect lines of words, not isolated numbers. Regarding AI, Audiveris current neural network is a shallow one. Using deep learning techniques, I developed a patch classifier last year, able to detect and classify a potential music symbol around any given point of the input image. It worked but was too slow, the processing of a full page via a sliding window would have taken several days! :-) |
Beta Was this translation helpful? Give feedback.
I was used to Marcel Dadi's books of good quality with fret + stems + beams.
But tablatures can be very different from one editor to the other.
This the reason why I'm asking you to provide pointers to the kinds of tablatures you are aiming at.
And then, we'll be able to evaluate the global feasability of the recognition task.
The current Audiveris version uses various methods to process a partition, depending on the items to handle. For instance, note heads use template matching, beams use image closing, texts use an external OCR software, stand alone symbols use a neural network trained on representative glyphs, etc.
Regarding tablatures, I think OCR would not be the key component, beca…