A modern Conformer ASR model's predictions are tied to vowel formants (F1 and F2), sibilant fricative spectral peaks, and plosive release bursts, per SPES feature attributions on TIMIT.
This is the first in-depth analysis of saliency maps in relation to fine-grained acoustic patterns across three phoneme classes
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
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Echoes of Phonetics: Unveiling Relevant Acoustic Cues for ASR via Feature Attribution
A modern Conformer ASR model's predictions are tied to vowel formants (F1 and F2), sibilant fricative spectral peaks, and plosive release bursts, per SPES feature attributions on TIMIT.