Using the model's own encoder as a feature loss improves perceptual quality and iterative stability of a speech enhancement model compared with a WavLM-based loss.
We also thank Sai Dhawal Phaye for discussions during the early stages of MAL, and Kanav Sabharwal for his feedback on the writing
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
-
Model as Loss: A Self-Consistent Training Paradigm
Using the model's own encoder as a feature loss improves perceptual quality and iterative stability of a speech enhancement model compared with a WavLM-based loss.