Automatic Speech Recognition
NeMo
ONNX
Safetensors
GGUF
parakeet_tdt
parakeet
tdt
sherpa-onnx
multilingual
speech-recognition
gabor
fastconformer
Instructions to use oruk/orukeet with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- NeMo
How to use oruk/orukeet with NeMo:
import nemo.collections.asr as nemo_asr asr_model = nemo_asr.models.ASRModel.from_pretrained("oruk/orukeet") transcriptions = asr_model.transcribe(["file.wav"]) - Notebooks
- Google Colab
- Kaggle
Publish opt-in iOS INT8 integration metadata and device qualification scope
Browse files- coreml/README.md +33 -3
- coreml/SHA256SUMS.txt +1 -0
- coreml/manifest.json +20 -0
coreml/README.md
CHANGED
|
@@ -1,6 +1,36 @@
|
|
| 1 |
-
# Orukeet r3 Core ML
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
-
Portable Core ML bundles for Apple Silicon, macOS 14+, and **FluidAudio 0.15.5**.
|
| 4 |
These are the same archives as the
|
| 5 |
[GitHub preview release](https://github.com/Oruk-AI/orukeet/releases/tag/coreml-taptalk-preview-20260915),
|
| 6 |
published here for TapTalk and other applications to download.
|
|
@@ -87,7 +117,7 @@ FLEURS recordings across eight languages. Core ML WER was 7.64%, compared with
|
|
| 87 |
|
| 88 |
The baseline uses 6-bit LUT/FP16 weights; the greedy profile removes unused
|
| 89 |
top-64 joint calculations. It is unsuitable for decoding paths that require
|
| 90 |
-
those outputs. The experimental precision profile is
|
| 91 |
|
| 92 |
This is a preview. Other Macs, older OS versions, power use, and the full
|
| 93 |
25-language Core ML evaluation remain unqualified. FluidAudio's default
|
|
|
|
| 1 |
+
# Orukeet r3 Core ML previews
|
| 2 |
+
|
| 3 |
+
Portable Core ML bundles for Apple Silicon. The original macOS previews below use **FluidAudio 0.15.5**.
|
| 4 |
+
The iOS 17+ opt-in candidate uses the pinned Oruk FluidAudio fork through OrukeetCoreML.
|
| 5 |
+
|
| 6 |
+
## iOS / OpenWhispr candidate
|
| 7 |
+
|
| 8 |
+
The experimental INT8 encoder archive is published for opt-in integration with
|
| 9 |
+
[OrukeetCoreML v0.1.2-coreml.1](https://github.com/Oruk-AI/orukeet/releases/tag/v0.1.2-coreml.1). Audio is 16 kHz mono;
|
| 10 |
+
record first, then run local batch inference. Keep the engine loaded between recordings.
|
| 11 |
+
|
| 12 |
+
- [Download the INT8 archive](https://huggingface.co/oruk/orukeet/resolve/419d7f79e290127e202a0f610509868d314743eb/coreml/orukeet-r3-coreml-int8sym-encoder-only-experimental-20260920-deflated.zip?download=true)
|
| 13 |
+
- Immutable model revision: `419d7f79e290127e202a0f610509868d314743eb`
|
| 14 |
+
- Download bytes: **554,985,744**
|
| 15 |
+
- SHA-256: `24df9ff76f00f86f9ae1fd601cbbcab1d1eac98c7e8107de67444a7858d88b8b`
|
| 16 |
+
- Archive root: `orukeet-r3-coreml-int8sym-encoder-only-experimental-20260920/`
|
| 17 |
+
|
| 18 |
+
Use the SDK model store to authenticate, safely extract and compile packages on-device.
|
| 19 |
+
It retains the authenticated source ZIP for offline recompilation after an OS update.
|
| 20 |
+
The installed cache is approximately **1.19 GB**, including the retained ZIP; first-install
|
| 21 |
+
peak is approximately **1.82 GB** before an app-specific free-space margin. Compilation size
|
| 22 |
+
varies by device and OS. Legacy installs without a retained source require one fresh download.
|
| 23 |
+
|
| 24 |
+
This candidate is experimental and opt-in. The original 400-clip diagnostic panel does not
|
| 25 |
+
establish 25-language production acceptance; it contains language/script regressions, including
|
| 26 |
+
Slovenian Cyrillic output. Full-corpus qualification is in progress, and physical iPhone
|
| 27 |
+
performance/thermal/microphone acceptance remains required. The archive itself is immutable;
|
| 28 |
+
qualification updates are separate metadata, not modifications of its weights or decoder.
|
| 29 |
+
|
| 30 |
+
[Integration and device acceptance guide](https://github.com/Oruk-AI/orukeet/blob/v0.1.2-coreml.1/integrations/openwhispr/ios/README.md)
|
| 31 |
+
|
| 32 |
+
## Original macOS previews
|
| 33 |
|
|
|
|
| 34 |
These are the same archives as the
|
| 35 |
[GitHub preview release](https://github.com/Oruk-AI/orukeet/releases/tag/coreml-taptalk-preview-20260915),
|
| 36 |
published here for TapTalk and other applications to download.
|
|
|
|
| 117 |
|
| 118 |
The baseline uses 6-bit LUT/FP16 weights; the greedy profile removes unused
|
| 119 |
top-64 joint calculations. It is unsuitable for decoding paths that require
|
| 120 |
+
those outputs. The separate iOS experimental precision profile is described above.
|
| 121 |
|
| 122 |
This is a preview. Other Macs, older OS versions, power use, and the full
|
| 123 |
25-language Core ML evaluation remain unqualified. FluidAudio's default
|
coreml/SHA256SUMS.txt
CHANGED
|
@@ -1,2 +1,3 @@
|
|
| 1 |
beccdc6f18c4b10527a764f6e3ab12e3e11b969220c0cee175b3bb7eaa94290e orukeet-r3-coreml-greedy.zip
|
| 2 |
b2a6efc4ed3280c860f29b3e2e2ea242ade14c6482c94f1c8d3e8551d5edb626 orukeet-r3-coreml-baseline.zip
|
|
|
|
|
|
| 1 |
beccdc6f18c4b10527a764f6e3ab12e3e11b969220c0cee175b3bb7eaa94290e orukeet-r3-coreml-greedy.zip
|
| 2 |
b2a6efc4ed3280c860f29b3e2e2ea242ade14c6482c94f1c8d3e8551d5edb626 orukeet-r3-coreml-baseline.zip
|
| 3 |
+
24df9ff76f00f86f9ae1fd601cbbcab1d1eac98c7e8107de67444a7858d88b8b orukeet-r3-coreml-int8sym-encoder-only-experimental-20260920-deflated.zip
|
coreml/manifest.json
CHANGED
|
@@ -23,6 +23,26 @@
|
|
| 23 |
"verified_files": 17,
|
| 24 |
"compiled_cache_included": false,
|
| 25 |
"archive_root": "orukeet-r3-coreml-baseline/"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
}
|
| 27 |
}
|
| 28 |
}
|
|
|
|
| 23 |
"verified_files": 17,
|
| 24 |
"compiled_cache_included": false,
|
| 25 |
"archive_root": "orukeet-r3-coreml-baseline/"
|
| 26 |
+
},
|
| 27 |
+
"int8sym_encoder_only_experimental": {
|
| 28 |
+
"filename": "orukeet-r3-coreml-int8sym-encoder-only-experimental-20260920-deflated.zip",
|
| 29 |
+
"bytes": 554985744,
|
| 30 |
+
"sha256": "24df9ff76f00f86f9ae1fd601cbbcab1d1eac98c7e8107de67444a7858d88b8b",
|
| 31 |
+
"revision": "419d7f79e290127e202a0f610509868d314743eb",
|
| 32 |
+
"compiled_cache_included": false,
|
| 33 |
+
"archive_root": "orukeet-r3-coreml-int8sym-encoder-only-experimental-20260920/",
|
| 34 |
+
"status": "experimental_opt_in",
|
| 35 |
+
"minimum_ios": "17.0",
|
| 36 |
+
"fluid_audio_tag": "0.15.5-orukeet.1",
|
| 37 |
+
"fluid_audio_commit": "ddc95f4d03d5be12bf84eeb6e4bde5b724356d88",
|
| 38 |
+
"source_pr": "https://github.com/Oruk-AI/orukeet/pull/10",
|
| 39 |
+
"qualification": {
|
| 40 |
+
"full_fleurs": "in_progress",
|
| 41 |
+
"physical_iphone": "pending"
|
| 42 |
+
},
|
| 43 |
+
"source_code_revision": "6c37c587fabcef8b0e932584fbf22a79fcb9d89e",
|
| 44 |
+
"sdk_release_tag": "v0.1.2-coreml.1",
|
| 45 |
+
"github_release": "https://github.com/Oruk-AI/orukeet/releases/tag/v0.1.2-coreml.1"
|
| 46 |
}
|
| 47 |
}
|
| 48 |
}
|