Beyond Answer Changes: Matched Same-Class Controls for Local Audio Evidence
For audio-language models, an answer change after a local sound-event edit can reflect a response to the event's class, sensitivity to waveform rewriting, or both. We compare matched same-class and different-class replacements of a queried event. Both use the same interval, level estimate, donor fitting, and crossfade, while the same-class replacement preserves the correct answer. Across 600…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.