Breeze ASR 25 is fine-tuned from Whisper large-v2 with significantly better performance on Taiwanese Mandarin and Mandarin-English code-switching — up to 56% lower WER compared to the original Whisper large-v2.
This repo provides the 4-bit palettized CoreML version with outlier decomposition, ready for iPhone, iPad, Mac, and potentially Vision Pro.
1importWhisperKit23// Download and extract model files to a local directory4let config =WhisperKitConfig(5 modelFolder:"/path/to/breeze-asr-25-whisperkit-coreml",6 download:false7)8let whisperKit =tryawaitWhisperKit(config)9let result =tryawait whisperKit.transcribe(audioPath:"audio.wav")10print(result.text)
Not real-time: At ~1 GB, this model is best for batch transcription, not live dictation. For real-time keyboard use, consider smaller models (whisper-base or whisper-small).
First load time: ANE compilation on first use takes ~10-12 minutes per component. Subsequent loads are cached.
visionOS: CoreML is available on visionOS but WhisperKit visionOS support needs verification.
License
This model is released under the Apache 2.0 License, same as the original Breeze ASR 25 model.
If you use this model, please cite the original Breeze ASR 25 paper:
bibtex
1@article{chou2025selfrefiningframeworkenhancingasr,
2 title={A Self-Refining Framework for Enhancing ASR Using TTS-Synthesized Data},
3 author={Cheng Kang Chou and Chan-Jan Hsu and Ho-Lam Chung and Liang-Hsuan Tseng and Hsi-Chun Cheng and Yu-Kuan Fu and Kuan Po Huang and Hung-Yi Lee},
4 journal={arXiv preprint arXiv:2506.11130},
5 year={2025}
6}
