Orukeet
A 25-language speech recognition model built from NVIDIA's Parakeet TDT 0.6B v3, replacing half its encoder's temporal filters with fitted, frozen Gabor kernels and retraining the rest, reporting lower error rates than Parakeet on most benchmarks.
Orukeet is a 25-language automatic speech recognizer built from NVIDIA Parakeet TDT 0.6B v3. It replaces half of the encoder's temporal depthwise filters with 12,288 fitted, frozen Gabor kernels and trains the remaining parameters on multilingual and multi-accent data.
It reports outperforming Parakeet on 61 of 74 tested splits, including a lower word-error rate on LibriSpeech test-clean and test-other and on the English portion of FLEURS, and a 10.6 percent relative reduction in pooled word-error rate across all 25 FLEURS languages. NeMo, ONNX INT8, and native Q8 and F16 builds all derive from the same released checkpoint.
It is available as a local model inside the OpenWhispr and TapTalk dictation apps, and can be run directly with a supplied Python tutorial and package for local deployment, using Python 3.12 or newer in a virtual environment.