Learning Distributed Representations for Multiple-Viewpoint Melodic Prediction

The analysis of sequences is important for extracting in- formation from music owing to its fundamentally temporal nature. In this paper, we present a distributed model based on the Restricted Boltzmann Machine (RBM) for learning melodic sequences. The model is similar to a previous suc- cessful neural network model for natural language [2]. It is first trained to predict the next pitch in a given pitch se- quence, and then extended to also make use of information in sequences of note-durations in monophonic melodies on the same task. In doing so, we also propose an efficient way of representing this additional information that takes advantage of the RBM’s structure. Results show that this RBM-based prediction model performs better than previ- ously evaluated n-gram models and also outperforms them in certain cases. It is able to make use of information present in longer sequences more effectively than n-gram models, while scaling linearly in the number of free pa- rameters required.