What is source separation (Demucs, Spleeter)?

Source separation is a task in audio processing aimed at separating a mixed audio signal into individual components or sound sources. For example, isolating vocals, bass, drums, and other instruments from a song.

Demucs and Spleeter are modern models and tools for source separation based on deep neural networks. They are trained on large datasets to effectively split audio tracks into their constituent parts.

Applications:

• Creating karaoke (removing vocals).
• Remixes and music production.
• Music analysis and recognition.

Technically, these models take a mixed audio signal as input and output several audio tracks, each corresponding to a separate sound source. They utilize architectures such as convolutional and recurrent neural networks, as well as transformers, to analyze temporal and spectral characteristics of sound.