A new automatic speech recognition (ASR) model named Orukeet has been released, built upon the NVIDIA Parakeet TDT 0.6B v3 architecture. Orukeet incorporates 12,288 fitted, frozen Gabor kernels and is trained on multilingual and multi-accent data, supporting 25 languages. Benchmarks indicate Orukeet surpasses Parakeet in performance across 61 out of 74 tested splits, showing improvements in Word Error Rate (WER) on datasets like LibriSpeech and FLEURS. AI
IMPACT Offers improved speech recognition performance, potentially benefiting applications relying on ASR.
RANK_REASON New model release with benchmark performance data. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →