Optical character recognition (OCR), a well-established digitization technique, is commonly used to convert the text in scanned documents into a searchable and editable form on the computer. But it can’t digitize other documents such as musical manuscripts and other types of manuscripts. A new approach developed by a team at Bina Nusantara University, situated in Jakarta, Indonesia, uses deep machine learning and a convolutional neural network which is trained to recognize the nuance of musical notation written on the manuscripts.
The system requires the clef, stave, and musical key to be in position, but these can easily be assigned in a template. When converting a scanned manuscript, it detects the position of each note on the stave to define the pitch. The next step uses a parallel algorithm to detect the duration of each note and identify the position of rests, silences, and other similar features in a manuscript. Once completely digitized, it is a trivial matter with current software to “play” the manuscript with all possible instrumental sounds on the computer or to even correlate a lyrical score with the music and let the computer sing the song. Scientists believe that once matured, OMR would have many applications in music performance, music education, and in archiving musical manuscript archives. The team suggests that their approach could allow software “application” developers to write a program for smartphones or tablets that would allow anyone, for instance, to quickly scan a score and do OMR on that manuscript.