A centroid-based speaker consistency loss plus conditional loss suppression improves target speaker extraction quality and speaker similarity across multiple backbone models.
Dataset We use the clean Libri2Mix [26] dataset with two-speaker mix- tures
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Enhancing Target Speaker Extraction with Explicit Speaker Consistency Modeling
A centroid-based speaker consistency loss plus conditional loss suppression improves target speaker extraction quality and speaker similarity across multiple backbone models.