Anomalous Note Change Detection of Unknown Monophonic Melodies

Abstract

Music melody is a sequence of music notes which are arranged in a musically satisfied manner. There should be a scale for each melody performance. A particular music scale has a set of ‘related notes’, and therefore, a melody consists of a set of scale satisfied notes. However, when ‘the scale un-related notes’ occur in the note sequence, it will provide less pleasant melodies. These ‘uncommon’ notes are the situations which refer as the ‘anomalous note’ in a particular melody. It is a major concern in the context of ‘melody evaluation’. In this study, a novel approach is proposed to detect anomalous notes changes of musical melodies. The proposed model is focused on to have two phases. Within the first phase, melodies are processed to have their pitch estimations. The steps of feature extraction and fundamental frequency estimations are involved in the first phase. After the pitch estimation, a note event model is employed with the application of Long Short Term Memory (LSTM) neural network for the detection of ‘anomalous note changes’ regarding the estimated pitch values of sampled melody signals. The proposed model is designed to focus on unknown monophonic melodies, which is the simplest type of musical texture and was able to have 69% overall accuracy for the used dataset.