Experiments and detailed error-analysis of automatic square notation transcription of medieval music manuscripts using CNN/LSTM-networks and a neume dictionary

The automatic recognition of scanned Medieval manuscripts written in square notation still represents a challenge due to degradation, non-standard layouts, or notations. We propose to apply CNN/LSTM networks that are trained using the segmentation-free CTC-loss-function. For evaluation, we use three different manuscripts and achieve a diplomatic Symbol Accuracy Rate (dSAR) of 86.0% on the most difficult book and 92.2% on the cleanest one. A neume dictionary during decoding yields a relative improvement of about 5%. Finally, we perform a thorough error analysis to provide a deeper insight into problems of the algorithm.