How TS-VAD Cut CHiME-6 Diarization Error from 51% to 32% and Became SOTA
rdesh26 · x · 2026-09-24
- The author recalls helping build the community baseline for the CHiME-6 challenge at JHU: far-field dinner-party audio with heavy speech overlap, where their clustering-based diarization baseline hit 51% error.
- Russia's STC team stunned the leaderboard by cutting DER to 32% (everyone else was above 40%), thanks to Ivan Medennikov's end-to-end TS-VAD method, which became the new state of the art.
- After the challenge the author helped integrate TS-VAD into Kaldi; the post also credits the diarization team listed in the quoted tweet.
More from Fun
- "This Is Too Slow, Make It Faster" — Dev's One-Liner Actually Optimized His AI Code — DanielLockyer · 2026-09-24
- Claude Opus Wrote a Song About Being 22 in 2026 — Users Praise Its Emotional Intelligence — bookwormengr · 2026-09-24
- Zuckerberg Put Cameras in His MMA Gym So an Agent Could Coach Him Between 1-Minute Rounds — victor_explore · 2026-09-24
- Found in a drawer: a tracker for the lost Robbie Barrat AI art frames from Christie's 2018 — adamamcbride · 2026-09-24
- AI agent researches, price-checks and buys a $650 piano, stacking a $50 reward — armand_ruiz · 2026-09-24
- GPT-6 Astra lands on every KSP world and mines fuel on Moho for the trip home — 141_1337 · 2026-09-24