---
license: apache-2.0
library_name: whisper.cpp
tags:
  - whisper
  - whisper.cpp
  - ggml
  - coreml
  - automatic-speech-recognition
  - speech-recognition
---

# ggml-breeze-asr-26

This repository contains `whisper.cpp`-compatible conversion artifacts for [`MediaTek-Research/Breeze-ASR-26`](https://huggingface.co/MediaTek-Research/Breeze-ASR-26).

Included files:

- `ggml-breeze-asr-26.bin`: ggml model for `whisper.cpp`
- `ggml-breeze-asr-26-encoder.mlmodelc/`: Core ML encoder compiled for Apple platforms

## Upstream model

- Base architecture: `openai/whisper-large-v2`
- Upstream fine-tuned model: `MediaTek-Research/Breeze-ASR-26`
- Original license: Apache-2.0

## Usage with whisper.cpp

Build `whisper.cpp` with Core ML enabled:

```bash
cmake -B build -DWHISPER_COREML=1
cmake --build build -j --config Release
```

Run transcription with the converted model:

```bash
./build/bin/whisper-cli -m models/ggml-breeze-asr-26.bin -f samples/jfk.wav
```

When the `.mlmodelc` directory is placed next to the `.bin` file, `whisper.cpp` will load it automatically on Apple platforms.

## Conversion

These artifacts were generated with `ggml-org/whisper.cpp` using:

```bash
./models/convert-hf-whisper-model.sh MediaTek-Research/Breeze-ASR-26
```