Mask polarization restores bimodality in SE model predictions via Wasserstein distance at test time, delivering consistent gains across domain shifts and architectures.
The most common approach identifies similar source sample to use as pseudo-labels [5, 6], but requires access to source data, defying the TTA paradigm
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Test-Time Adaptation For Speech Enhancement Via Mask Polarization
Mask polarization restores bimodality in SE model predictions via Wasserstein distance at test time, delivering consistent gains across domain shifts and architectures.