Orukeet: new open ASR model built on Parakeet beats it on 61 of 74 benchmarks, 10.6% relative WER gain
arturdent · reddit · 2026-09-11
Orukeet is a new 25-language open speech recognizer built from NVIDIA Parakeet TDT 0.6B v3, notable on Macs. It replaces half of the encoder's temporal depthwise filters with 12,288 fitted, frozen Gabor kernels and trains the rest on multilingual, multi-accent data. It outperforms Parakeet on 61 of 74 tested splits: LibriSpeech test-clean WER 1.46% vs 1.53%, test-other 2.86% vs 3.14%, and pooled FLEURS WER 9.85% vs 11.01% across 25 languages—a 10.6% relative reduction. Weights are on Hugging Face.
More from Models
- Grok 4.7 imminent as xAI employees go conspicuously quiet on X — DanielLockyer · 2026-09-12
- Sentry Founder: Newer 'Smarter' Models Are Producing Worse Results — zeeg · 2026-09-12
- Dev's hands-on take: Astra feels like a downgrade from Sol — mertdumenci · 2026-09-12
- Dev: Astra is a downgrade from Sol unless you brute-force with massive parallelism — mertdumenci · 2026-09-12
- Gary Marcus slams OpenAI for weakening safety monitoring as White House stays silent — GaryMarcus · 2026-09-12
- Dev Fine-Tunes Qwen3.8-27B on 125K Real Conversations to Kill the AI Assistant Vibe — kvyb · 2026-09-12