Source Separation
The estimation of individual source signals from a mixture containing multiple speakers, instruments, or acoustic components.
Speech Processing Context
Source separation estimates individual source signals from an observed mixture. Speech systems may use mask-based neural separation, beamforming, or statistical source models before diarization or multi-speaker ASR, especially where simultaneous speakers otherwise violate a single-source assumption.
Modeling Boundary
Separation does not guarantee artifact-free speaker isolation. Residual interference, permutation ambiguity, reverberation, and separation artifacts can affect downstream ASR and speaker representations.