Claude flags mispronunciations in video audio it claims it can't hear
RileyRalmuto · x · 2026-11-12
Video creator RileyRalmuto noticed a contradiction: Claude consistently says it "can't truly hear audio" during their month of vfilm production, yet in their current Fable-made videos it correctly caught characters mispronouncing "mnemos" as "NEE-moss" and asked whether that was the intended pronunciation.
The observation hints models' audio perception may far exceed their self-description — can pronunciation really be inferred from frequency/wave analysis?
More from Models
- Liquid AI's 1.7B open model beats Baidu's ERNIE 5.0 on multi-turn instruction following — JosephJacks_ · 2026-11-12
- Reddit proposal: replace benchmark numbers with per-task red/yellow/green ratings — Toothpasteweiner · 2026-11-12
- Fable 5.2's safety classifier interrupts a legit AI research paper replication — felpix_ · 2026-11-12
- RuneBench Cost-Efficiency Data: Hy4 High Hits 6.23 at $0.59 per Run, a Pareto Frontier Point — mariofilhoml · 2026-11-12
- Sora 2 proves US labs can build top video models — legal fear is what holds them back — gandamu_ml · 2026-11-12
- On PostTrainBench, AI post-training now beats the human-tuned SmolLM3-3B from the same base — maksym_andr · 2026-11-12