A controlled benchmark of 13 pretrained audio models shows source direction, distance, and class are much more decodable than room size, shape, and reverb time.
Towards a unified representation eval- uation framework beyond downstream tasks,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Probing Spatial Structure in Pretrained Audio Representations
A controlled benchmark of 13 pretrained audio models shows source direction, distance, and class are much more decodable than room size, shape, and reverb time.