Kyutai's 100M PocketTTS trains with Kaiming He's Drifting, WER under 1%
serrjoa · x · 2026-10-06
- Kyutai labs released a technical blog on PocketTTS, a 100M-parameter on-device TTS model with voice cloning, achieving less than 1% WER.
- It's trained via Drifting, a recent one-step continuous generative objective from Deng et al. (Kaiming He's team). The team says it's the first speech model and the first autoregressive model trained this way.
- The reposter notes this is the first application of Drifting to autoregressive continuous generation.
More from Research
- MemAdapter uses counterfactual reasoning to curb memory-induced sycophancy in LLM agents — Ruqing Ning · 2026-10-06
- Peking University's Code2Games gets coding agents to build playable UE5 game worlds — PekingUniversity · 2026-10-06
- QuantCode: domain pretraining + SFT lifts Qwen trading-code pass from 27.8% to 58.2% — Alexey Chernysh · 2026-10-06
- Subsampling and extrapolation keep the Mandelbrot area estimate unbiased near the boundary — geoffreyirving · 2026-10-06
- New estimate sits 6.5e-9 below Hsing Lo's 2025 value; reproduction suggests the gap is a fluctuation — geoffreyirving · 2026-10-06
- How the new Mandelbrot area estimate works: quadtree pruning plus random boundary sampling on H200s — geoffreyirving · 2026-10-06