Melody Extraction

Melody extraction, the process of identifying the main melody line from audio recordings, particularly polyphonic music, is a core task in music information retrieval. Current research focuses on improving accuracy and robustness using deep learning models, often employing architectures like convolutional neural networks and recurrent neural networks, with refinements including specialized input representations (e.g., incorporating harmonic information) and loss functions designed to enhance stability and sensitivity. These advancements have significant implications for applications such as music transcription, query-by-humming systems, and music analysis, ultimately contributing to a deeper understanding of music perception and processing.