Automatic music generation using a bio-inspired algorithm-based deep learning model
V. Bhuvana Kumar, M. Kathiravan
Abstract
In recent years, automatic music generation has played a vital role in getting multimedia products cheaper and faster. For automatic music generation, both machine learning and deep learning methods were presented. The researchers, in particular, have used long short-term memory (LSTM). Although the LSTM model produces better results, its prediction accuracy for music generation needs to be improved further. Thus, an optimised LSTM model is presented for automatic music generation. Namely, to improve the efficiency of LSTM, an adaptive crocodile optimisation algorithm (ACOA) is presented. Using ACOA, the weight parameters of the LSTM are optimised. It leads to enhanced efficiency in music generation or music vector prediction. The proposed scheme is evaluated using a classical music musical instrument digital interface (MIDI) dataset. The paper's findings show that the proposed ACOA-LSTM outperforms the conventional LSTM in prediction accuracy.