Open-weight voice model Confucius4 survives screaming World Cup commentary stress test
dansuy_gaming · reddit · 2026-09-04
A voice-model enthusiast stress-tested the open-weight voice model Confucius4 with three brutal clips: a Spanish commentator screaming through a hat-trick call, an English commentator's held-breath-then-explosion moment, and a visibly shaken goalkeeper's post-match interview — all recently recorded World Cup commentary translated into other languages, with no scripts available.
These high-emotion clips typically expose cloned speech models (robotic screams or dead, context-free translation). Confucius4 takes the voice directly from audio rather than a transcript first. Results: short, high-emotion clips carried the trembling and emotion into the translation with little synthetic quality; longer sentences showed more synthetic artifacts.
More from Multimodal
- Voxtral 3 ships a LiveKit-based voice agent SDK streaming listen-think-speak — MistralAI · 2026-10-21
- Voxtral Mini TTS 3 hits SOTA expressiveness in 35+ languages with tag-based control — MistralAI · 2026-10-21
- Mistral launches Voxtral 3 speech models and new Voice Agents API — MistralAI · 2026-10-21
- Open-sourced pixel art normals generator ships as Rust CLI and wasm library — HugoDuprez · 2026-10-20
- Fibo 2.0 is strongest for ecommerce and creator content, weakest at UIUX design — ArtificialAnlys · 2026-10-20
- Fibo 2.0 leads old Fibo on all 9 capabilities, best at reasoning and anatomy — ArtificialAnlys · 2026-10-20