BuzzASR releases 102 monolingual ASR models, beating Whisper on 77 languages with SOTA on 27
LChoshen · x · 2026-09-15
BuzzASR takes a different approach from big labs: instead of one multilingual model, it ships a swarm of 102 monolingual speech recognition models, beating Whisper on 77 languages and reaching SOTA on 27.
The poster notes the project comes from a university with world-class ASR that lacks capacity to process all its own data — a missed opportunity for non-English speakers long underserved by single multilingual models.
More from Research
- TailSFT: Skipping SFT Gains Improves pass@16 and RL Exploration in GRPO — tw_killian · 2026-09-15
- STARK Co-inventor Notes STARKs Are Post-Quantum Secure — jamestagg · 2026-09-15
- Self-evolving AI agents shouldn't grade their own homework: a 7-step evidence-gated roadmap — MaryamMiradi · 2026-09-15
- I2T loss stays comparable across tokenizers and predicts image generation quality — peterxichen · 2026-09-15
- Fruit fly brain with 166,700 neurons installed in a physical robot that walks — udmrzn · 2026-09-15
- Image Tokenizers Define the Visual Language of Unified Multimodal Models — peterxichen · 2026-09-15