Music Demixing

Music demixing, the separation of individual instruments or vocal tracks from a mixed audio recording, aims to improve audio quality and enable customized listening experiences. Current research focuses on developing robust deep learning models, including transformer-based architectures and variations of convolutional and recurrent neural networks, often employing frequency-domain processing and techniques like band-splitting to enhance separation accuracy. These advancements are evaluated through challenges using benchmark datasets and metrics like signal-to-distortion ratio (SDR), driving improvements in both the objective quality and perceptual experience of separated audio, with applications ranging from hearing aid technology to cinematic audio mastering.