Store › model › Speech
orukeet
by oruk · source Hugging Face · updated 2026-09-24
cc-by-sa-4.0705 MB~2 GB RAMsource aliveunlabeled
Nathan Roll1,2 · Irene Yi1,2 · Büşra Marşan1,2 Vianney Grenez1 · Gabriel Stein4 · Momcilo Mrkaic5 Pavle Padjin5 · Vladimir Zeljkovic5 · Calbert Graham1,3
Add to LogiShell Open in the web IDE
The button opens LogiShell with this card; nothing installs from a link by itself. Inside the app the install goes through lsh models install hf:oruk/orukeet and its progress lives in the Resource Center.
Source and license
- Source: https://huggingface.co/oruk/orukeet
- License: cc-by-sa-4.0
- Requirements: about 2 GB of RAM, 705 MB on disk (estimate: suggested file size × 1.15 + 0.5 GB; a real measurement comes with lsh models). Runs with whisper.cpp.
- Tags:
nemoonnxsafetensorsggufparakeet_tdtparakeettdtsherpa-onnxmultilingualspeech-recognitiongaborfastconformerautomatic-speech-recognitionbghrcs
Numbers
- 40,095 downloads on Hugging Face
- 99 likes
- license cc-by-sa-4.0
- 0.7 GB for orukeet-transcribe-cpp-Q8_0.gguf
- 54,249 stars on ggml-org/whisper.cpp
- 352 open issues and PRs
- last release v1.9.5 on 2026-10-06
- 15,664 npm downloads a week for nodejs-whisper
- latest nodejs-whisper@0.3.1
Numbers as of 2026-10-09 19:01 UTC, from the source API.
Summary
Summary not ready yet: the numbers are here, the text is not. It is written by the collector through the LogiShell model facade when a provider key is present.
Reviews
No reviews yet. Reviews are written inside LogiShell: open this card in the app.
Files
| file | quant | size |
|---|---|---|
model.safetensors | 2.3 GB | |
onnx/combined-v0.1.0-int8/decoder_joint-model.int8.onnx | 17 MB | |
onnx/combined-v0.1.0-int8/encoder-model.int8.onnx | 623 MB | |
onnx/combined-v0.1.0-int8/nemo128.onnx | 136 KB | |
onnx/sherpa-v0.1.0-int8/decoder.int8.onnx | 11 MB | |
onnx/sherpa-v0.1.0-int8/encoder.int8.onnx | 623 MB | |
onnx/sherpa-v0.1.0-int8/joiner.int8.onnx | 6 MB | |
orukeet-transcribe-cpp-Q8_0.gguf | Q8_0 | 705 MB |
orukeet-v0.1.0-f16.gguf | F16 | 1.2 GB |
orukeet-v0.1.0-q8.gguf | 681 MB | |
transcribe-cpp/orukeet-Q8_0.gguf | Q8_0 | 705 MB |
From the source README
Orukeet
Nathan Roll1,2 · Irene Yi1,2 · Büşra Marşan1,2
Vianney Grenez1 · Gabriel Stein4 · Momcilo Mrkaic5
Pavle Padjin5 · Vladimir Zeljkovic5 · Calbert Graham1,3
1 Oruk AI
2 Stanford University
3 University of Cambridge
4 OpenWhispr
5 Hoid
Orukeet is a 25-language 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.
Orukeet outperforms Parakeet on 61 of 74 tested splits, including LibriSpeech test-clean (1.46% vs. 1.53% WER), test-other (2.86% vs. 3.14%), and FLEURS English (3.82% vs. 4.28%). Across all 25 FLEURS languages, pooled WER is 9.85% vs. 11.01%, a 10.6% relative reduction. Final adaptation and checkpoint selection use LibriSpeech test-other.
Use Orukeet for recordings, media, batch transcription, server workers and interactive applications. NeMo, ONNX INT8, native Q8 and native F16 all derive from the same r3 release checkpoint (`031c8ddab484`).
Code · OpenWhispr PR · Technical report · Artifact hashes
Run Orukeet with Transformers
A standard FP32 Transformers export is available at the repository root. It uses
`ParakeetForTDT` without custom remote code and works with Buzz's existing Hugging
Face model option. See setup, conversion provenance and runtime qualification.
The NeMo evaluation below remains the source-model benchmark; the Transformers
export has separate compatibility measurements.
Run Orukeet with NeMo
Use a CUDA-enabled PyTorch environment with `nemo_toolkit[asr]==3.0.0` and `huggingface-hub`. The [recorded source environment](https://github.com/Oruk-AI/orukeet/blob/main/evidence/standard-asr-20260908/r…
Card id model:hf:oruk/orukeet · collected 2026-10-09 19:01 UTC · JSON