Classical Music Generation using LSTM-based Models

AI generated music has witnessed a lot of innovative methods in the past. Most of the recent generation techniques revolve around machine learning, especially deep learning. This paper explores music generation using Long Short-Term Memory (LSTMs), which are wellsuited for handling sequential data. Since music typically consists of a sequential set of notes and chords, LSTMs can be used to extract important features such as notes, duration, and offset present in the input music. Considering these features is essential for generating a smooth and coherent output. A note accuracy of 86.7% and training loss of 0.56 was achieved with the method described in this paper. Thus, the ability of LSTMs to capture distinct features in the music, and to generate new music based on these features is investigated.