| license: mit | |
| language: | |
| - en | |
| - multilingual | |
| tags: | |
| - whisper | |
| - coreml | |
| - apple-silicon | |
| - neural-engine | |
| - speech-recognition | |
| - asr | |
| library_name: whisper.cpp | |
| # Whisper CoreML Models | |
| Pre-converted CoreML encoder models for [whisper.cpp](https://github.com/ggerganov/whisper.cpp), optimized for Apple Neural Engine (ANE) acceleration. | |
| ## Models | |
| | Model | Size | Description | | |
| |-------|------|-------------| | |
| | `ggml-large-v3-turbo-encoder.mlmodelc` | ~1.3 GB | Whisper large-v3-turbo encoder, ANE-optimized | | |
| ## Usage | |
| These models are used by [whisper-coreml](https://github.com/sebastian-software/whisper-coreml) for Node.js speech recognition with Neural Engine acceleration. | |
| ```bash | |
| npm install whisper-coreml | |
| npx whisper-coreml download | |
| ``` | |
| ## Conversion | |
| Models were converted using whisper.cpp's conversion tools: | |
| ```bash | |
| python3 models/convert-whisper-to-coreml.py --model large-v3-turbo --encoder-only True --optimize-ane True | |
| xcrun coremlc compile models/coreml-encoder-large-v3-turbo.mlpackage models/ | |
| ``` | |
| ## Performance | |
| On Apple M1 Ultra: | |
| - **14x real-time** transcription speed | |
| - 1 hour of audio in ~4.5 minutes | |
| ## License | |
| MIT - Same as whisper.cpp | |
| ## Credits | |
| - [OpenAI Whisper](https://github.com/openai/whisper) | |
| - [whisper.cpp](https://github.com/ggerganov/whisper.cpp) | |
| - [Apple Neural Engine Transformers](https://github.com/apple/ml-ane-transformers) | |