# CIViC: Evidence Gates for Audio-Conditioned Visual Prediction on Proxy Tasks

arXiv: https://arxiv.org/abs/2610.15327 · HTML: https://mlsanity.com/paper/2610.15327/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 11 of its 17 references that are in the map.
- First posted on arXiv 11 Oct 2026.

## Builds on (most cited first, 11 of 11)

- [AVA-ActiveSpeaker: An Audio-Visual Dataset for Active Speaker Detection](https://mlsanity.com/paper/1901.01342/) (2019)
- [Let Them Talk: Audio-Driven Multi-Person Conversational Video Generation](https://mlsanity.com/paper/2505.22647/) (2025)
- [VLOGGER: Multimodal Diffusion for Embodied Avatar Synthesis](https://mlsanity.com/paper/2403.08764/) (2024)
- [DualTalk: Dual-Speaker Interaction for 3D Talking Head Conversations](https://mlsanity.com/paper/2505.18096/) (2025)
- [SpeakerVid-5M: A Large-Scale High-Quality Dataset for Audio-Visual Dyadic Interactive Human Generation](https://mlsanity.com/paper/2507.09862/) (2025)
- [Animating Face using Disentangled Audio Representations](https://mlsanity.com/paper/1910.00726/) (2019)
- [AnyTalker: Scaling Multi-Person Talking Video Generation with Interactivity Refinement](https://mlsanity.com/paper/2511.23475/) (2025)
- [Audio-Sync Video Generation with Multi-Stream Temporal Control](https://mlsanity.com/paper/2506.08003/) (2025)
- [UniLS: End-to-End Audio-Driven Avatars for Unified Listening and Speaking](https://mlsanity.com/paper/2512.09327/) (2025)
- [Silence is Golden: Leveraging Adversarial Examples to Nullify Audio Control in LDM-based Talking-Head Generation](https://mlsanity.com/paper/2506.01591/) (2025)
- [Assessing Identity Leakage in Talking Face Generation: Metrics and Evaluation Framework](https://mlsanity.com/paper/2511.08613/) (2025)

## Abstract (arXiv)

Removing audio can shift a visual predictor's output without showing whether the task uses the intended audio information. We introduce Conditional Input and Visual Interpretation Checks (CIViC), a four-gate protocol that links an interpretation to an intervention, held-out task comparison, or endpoint control. We demonstrate CIViC on actor-disjoint face-raster tests, a synthetic audio-feature mixture, and source-video-disjoint SpeakerVid proxy tasks. In the static face-raster test, cross-actor audio replacement meets the registered detection rule with a small effect; on a separate temporal endpoint, constant-audio fits score better than audio-trained fits in four matched seeds. In the synthetic mixture, the condition-augmented fit has higher 1,000-step RMSE than the audio-only fit (0.1807 vs. 0.1707), with the gap shrinking during training. In the SpeakerVid proxy, the five-feature-zero fit (five low-level audio summaries zeroed; identity, dialogue, and condition inputs retained) has lower whole-frame grayscale RMSE than aligned audio across four held-out conditions and five paired seeds, while aligned audio has lower segment-drift error. The results show that output activation, held-out error, and endpoint controls support different interpretations across these proxy tasks.