Reproduce “Echoes of Phonetics: Unveiling Relevant Acoustic Cues for ASR via Feature Attribution” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dennis Fucci, Marco Gaido, Matteo Negri, Mauro Cettolo, Luisa Bentivogli
The authors test whether a speech recognizer’s salient regions align with phonetic cues. They perturb spectrogram groups and measure effects on predicted tokens. In correctly transcribed TIMIT SX examples, saliency generally covers vowel spans and aligns more with F1 and F2; sibilants align more clearly than non-sibilant fricatives, and plosive releases more than closures. This pattern for one model does not prove causal necessity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.