Breeze-ASR-25 GGML
GGML conversion of MediaTek-Research/Breeze-ASR-25 for use with whisper.cpp and compatible apps.
Model Details
- Base model: MediaTek-Research/Breeze-ASR-25
- Format: GGML (via whisper.cpp
convert-h5-to-ggml.py) - Language: Traditional Chinese / Taiwanese Mandarin
- License: Apache 2.0
Files
| File | Size | Description |
|---|---|---|
ggml-model.bin |
~2.9 GB | GGML model for whisper.cpp |
Usage
whisper.cpp
./main -m ggml-model.bin -f audio.wav -l zh
WhisperASR (macOS app)
WhisperASR is a native macOS app that uses this model. Download
ggml-model.bin and set the path in Settings.
Convert yourself
git clone https://github.com/ggml-org/whisper.cpp
git clone https://github.com/openai/whisper
pip install transformers numpy huggingface_hub
python whisper.cpp/models/convert-h5-to-ggml.py \
<path-to-MediaTek-Research/Breeze-ASR-25> \
<path-to-openai/whisper> \
<output-dir>
Attribution
Original model by MediaTek Research. Released under Apache 2.0.
Model tree for danielkao0421/Breeze-ASR-25-ggml
Base model
openai/whisper-large-v2 Finetuned
MediaTek-Research/Breeze-ASR-25