1. Calvo-Zaragoza, J.; Valero-Mas, J.J.; Pertusa, A

, vol. Proc. of International Society for Music Information Retrieval Conference (ISMIR), pp. . Suzhou, China (2017)

This work addresses the Optical Music Recognition (OMR) task in an end-to-end fashion using neural net- works. The proposed architecture is based on a Recurrent Convolutional Neural Network topology that takes as input an image of a monophonic score and retrieves a sequence of music symbols as output. In the first stage, a series of convolutional filters are trained to extract meaningful fea- tures of the input image, and then a recurrent block models the sequential nature of music. The system is trained us- ing a Connectionist Temporal Classification loss function, which avoids the need for a frame-by-frame alignment be- tween the image and the ground-truth music symbols. Ex- perimentation has been carried on a set of 90,000 synthetic monophonic music scores with more than 50 different pos- sible labels. Results obtained depict classification error rates around 2 % at symbol level, thus proving the po- tential of the proposed end-to-end architecture for OMR. The source code, dataset, and trained models are publicly released for reproducible research and future comparison purposes.