CNNs with 2D convolutions and frequency dynamic layers upsample 4-mic to 32-mic covariance matrices on STARSS23 data, lowering RMSE from 0.548 baseline to 0.432 and producing beamforming maps closer to 32-mic ground truth.
Deepwave: a recurrent neural-network for real-time acoustic imaging,
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
-
CNN Models for Microphone Array Covariance Matrix Upsampling and Acoustic Imaging
CNNs with 2D convolutions and frequency dynamic layers upsample 4-mic to 32-mic covariance matrices on STARSS23 data, lowering RMSE from 0.548 baseline to 0.432 and producing beamforming maps closer to 32-mic ground truth.