Instructions to use notchlive/whisperkit-coreml with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- WhisperKit
How to use notchlive/whisperkit-coreml with WhisperKit:
# Install CLI with Homebrew on macOS device brew install whisperkit-cli # View all available inference options whisperkit-cli transcribe --help # Download and run inference using whisper base model whisperkit-cli transcribe --audio-path /path/to/audio.mp3 # Or use your preferred model variant whisperkit-cli transcribe --model "large-v3" --model-prefix "distil" --audio-path /path/to/audio.mp3 --verbose
- Notebooks
- Google Colab
- Kaggle
NotchLive WhisperKit Core ML bundles
This repository provides the verified speech-model bundles used by NotchLive, a private, local-first speech workspace for Mac.
NotchLive normally downloads these files automatically. Manual downloads are intended for troubleshooting, verification, and distribution mirroring.
Published bundles
| Model | Archive | Size | SHA-256 |
|---|---|---|---|
| Tiny | bundles/tiny.notchlivemodel |
70,140,788 bytes | a3f7e9695c1040da2652a0cf7474acddbe88366533781b226b87907af479442d |
| Base | bundles/base.notchlivemodel |
134,089,083 bytes | feb0041900bfeb1785a5724a5c8f8f43e373ce39e9f12c9012d7d5c0d174bf29 |
| Small | bundles/small.notchlivemodel |
450,381,765 bytes | 4339ea684b9accaf4bc7c34ad3bf5c463c8ae26373e597162caf70bc45b3061c |
| Medium | bundles/medium.notchlivemodel |
1,424,991,213 bytes | 2dae20b0c8b638e68f181817ed5b2057d1aba60a50c91293784912c8f4ae4ae2 |
| Large Turbo (954 MB) | bundles/large-v3_turbo_954MB.notchlivemodel |
933,166,227 bytes | 8e4fcb29efaaed86f8134dfba5f774cee400af2017c2fe7f772873dbc2c953b8 |
| Large Turbo | bundles/large-v3_turbo.notchlivemodel |
2,964,827,414 bytes | b972aae0fc5cd4a76275c911999e511d103f5cdcfbd92af2daef8a19c9857810 |
The same metadata is available in catalog-v1.json, with conventional checksum lines in checksums.sha256.
What is inside each bundle
Each .notchlivemodel archive contains:
- pinned Core ML model files;
- matching tokenizer files;
- upstream revision identifiers;
- per-file SHA-256 hashes;
- the upstream OpenAI and Argmax license notices.
NotchLive verifies the archive checksum and every payload file before installation.
Intended use
These bundles are packaged for local speech recognition in NotchLive on macOS. After a model has been downloaded and installed, speech processing runs on the Mac. The initial model download still connects to the distribution host.
Model output can contain transcription errors, especially with unclear audio, overlapping speakers, unusual names, or unsupported language combinations. Users should review important transcripts before relying on them.
Provenance and licenses
NotchLive did not train these models and does not claim ownership of them:
- Whisper is developed by OpenAI and distributed under the MIT License.
- The Core ML conversions are provided by Argmax for WhisperKit and distributed under the MIT License.
The corresponding license notices are preserved inside every bundle.
Distribution mirrors
- Hugging Face:
notchlive/whisperkit-coreml - ModelScope:
NotchLive/whisperkit-coreml
The published .notchlivemodel artifacts are intended to remain byte-identical across both services.