A pseudo-supervised method that uses close-talk microphone recordings to estimate training targets lets a speech enhancement model adapt to real far-field data, cutting CER to 29.80% on MISP2023.
Front-end processing for the chime-5 dinner party scenario,
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
-
SuPseudo: A Pseudo-supervised Learning Method for Neural Speech Enhancement in Far-field Speech Recognition
A pseudo-supervised method that uses close-talk microphone recordings to estimate training targets lets a speech enhancement model adapt to real far-field data, cutting CER to 29.80% on MISP2023.