University of Tokyo researchers release Sidon AI for vocal restoration

Researchers at the University of Tokyo's Saruwata-Saito Laboratory have released Sidon, an AI audio restoration model capable of advanced voice reconstruction. While previous tools focused on 'subtraction'—removing noise or reverb—Sidon allows for the reconstruction of audio data that was never originally recorded.

Because Sidon was designed for speech, applying it directly to singing can distort pitch. To solve this, a method has been developed to combine the enhanced timbre provided by Sidon with the original pitch (F0) from the source recording. This approach enables the high-definition restoration of historical vocal recordings, such as century-old blues tracks, by effectively performing 'vocal super-resolution.'