Sony has advanced source separation using its proprietary AI technology, with practical applications expanding to film and anime remastering and improved headphone sound fields. Its recognition at Ars Electronica Festival 2026 and contribution to evala’s new work are also highlighted.
Sony began applying deep learning to source separation in 2013 and has improved its ability to separate audio sources by training models on guitar timbres and their frequency and temporal variations. The technology is now used in areas including film audio reconstruction and remastering, as well as AI beamforming for headphones.
For the 4K UHD editions of Lawrence of Arabia and Gandhi, source separation was used to extract dialogue and sound effects individually and reposition them in spatial audio, enhancing the sense of immersion. The technology has also been applied to anime remastering. References are made to the 4K remastered edition released in November 2025 and its presentation at events in 2026.
Sony has developed a new model for decomposing synthesizer timbres and put into practical use a method that separates elements such as melodies and timbres within a piece of music as clues. At Ars Electronica Festival 2026, the technology contributed to evala’s new work, Holy Echo.
Mitsufuji Yuki (光藤祐基) and others are working to bridge technology development and creative practice. In the future, the company aims to create an environment in which creators can take on new forms of expression, including restoring works and developing new forms of artistic expression. Sony says it will continue creating entertainment experiences through collaboration between technology and creators.