| license: apache-2.0 | |
| base_model: MediaTek-Research/Breeze-ASR-26 | |
| tags: | |
| - whisperkit | |
| - coreml | |
| - automatic-speech-recognition | |
| - taiwanese-hokkien | |
| # IPEVO WhisperKit CoreML Models | |
| WhisperKit-compatible Core ML models used by [Vurbo.ai](https://vurbo.ai). Converted with | |
| [argmaxinc/whisperkittools](https://github.com/argmaxinc/whisperkittools). | |
| ## Models | |
| | Folder | Base model | Precision | Size | Notes | | |
| |--------|-----------|-----------|------|-------| | |
| | `breeze-asr-26-q8` | [MediaTek-Research/Breeze-ASR-26](https://huggingface.co/MediaTek-Research/Breeze-ASR-26) | 8-bit palettized | ~1.5GB | Taiwanese Hokkien (台語) ASR, outputs Traditional Chinese | | |
| ## Attribution | |
| `breeze-asr-26-q8` is a derivative of Breeze-ASR-26 by MediaTek Research | |
| (Apache License 2.0), quantized to 8-bit and converted to Core ML for on-device | |
| inference via WhisperKit. `config.json` / `tokenizer.json` are from | |
| [openai/whisper-large-v2](https://huggingface.co/openai/whisper-large-v2) (same vocabulary). | |