---
title: "Beyond Answer Changes: Matched Same-Class Controls for Local Audio Evidence"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.37605v1"
permalink: "https://feed.astralor.com/items/itm_6410cba520f10dd1e73a35b0"
published: "2026-10-27T17:36:06.000Z"
score: 51
channel: "ai"
tags: ["RAG","多模态","Beyond Answer Changes","Matched Same-Class Controls","Local Audio Evidence","For","Both","Across"]
---

# Beyond Answer Changes: Matched Same-Class Controls for Local Audio Evidence

## At a glance

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 repla…

## Key points

- 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.

## Full text

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 synthetic soundscapes and three models from two architecture families, different-class replacements produce 49.7 to 60.9 percentage points more answer changes than same-class replacements. Same-class accuracy remains within 5.2 percentage points of original accuracy. Among trials with correct original and same-class answers, soundscape-weighted selective accuracy for the inserted class is 73.7 to 81.9%, averaging eligible trials within each soundscape and then weighting eligible soundscapes equally. This protocol tests how model answers respond to local changes in event class.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_6410cba520f10dd1e73a35b0
