Metric Analysis for Spatial Semantic Segmentation of Sound Scenes

Mayank Mishra, Paul Magron, Romain Serizel · 2025

Read Metric Analysis for Spatial Semantic Segmentation of Sound Scenes by Mayank Mishra, Paul Magron, Romain Serizel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-07075.txt.

Description

Mayank Mishra, Paul Magron, and Romain Serizel propose a fairer way to score systems that separate and classify sounds in a scene at once. This task, spatial semantic segmentation of sound scenes, is usually judged by separate separation and classification metrics, which makes systems hard to compare. The current joint metric can blur the two error types, because it matches sources using predicted labels and so can hide label swaps when separation itself is good.

Their metric, CASA-SDR, matches sources in a permutation-invariant, separation-first way before counting classification errors. Controlled tests with oracle separation and synthetic label errors show where the older metric over-penalizes, and comparisons on systems from the DCASE 2025 challenge follow. The new score gives a more interpretable, separation-centered assessment.

Catalog record

CREATORMayank Mishra, Paul Magron, Romain Serizel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-07075

Browse the full library · About the Tower of Babel