Handwritten music notation recognition using HMM — a non-gestural approach

We investigate the recognition of handwritten musical notation using Hidden Markov Models (HMMs). In a non-gestural approach, handwritten musical notation is entered naturally via a pen tablet as we would do using pen and paper. A sequence of observed ink patterns representing musical symbols is captured and used to construct different HMM models. The proposed approach exploits both global and local information derived from ink patterns which we have demonstrated the exploitation of this information via different features employed in different HMMs. The specificity and sensitivity measures of these different classification models are compared using unseen test sets. The experiment shows that using non-gestural method is a very good approach to obtain handwritten music notation input, as it is most natural and does not require user training. It is also shown that HMM is very suitable to be used as the classifier in this domain, showing very high recognition rates. Additionally, the experimental results also concluded that models from HMMs with more hidden states also outperform the HMM with a lesser number of hidden states since a larger model has more capacity. The results also suggest that HMMs offer flexibility in encoding useful knowledge in the models.