LSTM Based Monophonic Piano Melody Synthesis

Music composition has benefited from advancements in artificial intelligence (AI). This study explores AI-driven melody generation using machine learning techniques, specifically the Long Short-Term Memory (LSTM) network. The research follows three stages: Data pre-processing, model selection and training, and model evaluation. During pre-processing, songs with non-acceptable durations are filtered out, and the songs are transposed for compatibility and analysis. Symbolic representations, such as Musical Instrument Digital Interface (MIDI) notation, help identify patterns. Melodies are generated by feeding a seed sequence into the LSTM network, selecting the highest probability note iteratively. Decoding the representations produces a list of note and chord objects, forming coherent sheet music. Results demonstrate the potential of LSTM networks in producing diverse and high-quality music. Model evaluation includes loss function analysis and comparing the generated melodies with the original dataset. Future research should enhance model performance and address limitations. This study contributes to the understanding of AI in music composition, highlighting challenges, opportunities, and future potential.