A variational Bayesian framework that uses a trained neural network as a speech prior improves single-channel speech dereverberation and blind room impulse response estimation, with reported gains in MOS and WER.
High-order diffraction and dif- fuse reflections for interactive sound propagation in large environments,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
VINP: Variational Bayesian Inference with Neural Speech Prior for Joint ASR-Effective Speech Dereverberation and Blind RIR Identification
A variational Bayesian framework that uses a trained neural network as a speech prior improves single-channel speech dereverberation and blind room impulse response estimation, with reported gains in MOS and WER.