---
license: apache-2.0
language:
- nan
- zh
metrics:
- cer
base_model:
- MediaTek-Research/Breeze-ASR-26
pipeline_tag: automatic-speech-recognition
library_name: ctranslate2
tags:
- automatic-speech-recognition
- whisper
- faster-whisper
- ctranslate2
- taiwanese-hokkien
- taigi
- low-resource-language
- fine-tuned
---

# faster-whisper-Breeze-ASR-26

This is a [faster-whisper](https://github.com/SYSTRAN/faster-whisper) compatible conversion of [MediaTek-Research/Breeze-ASR-26](https://huggingface.co/MediaTek-Research/Breeze-ASR-26), converted to CTranslate2 format with **float16** quantization.

## Model Description

**BreezeASR-Taigi** is a Taiwanese Hokkien (Taigi / 台語) automatic speech recognition (ASR) model developed as part of the **Breeze Taigi** framework. It is fine-tuned from `openai/whisper-large-v2` on approximately **10,000 hours** of large-scale synthetic Taiwanese Hokkien speech data. The model transcribes spoken Taigi audio and outputs **Mandarin Chinese character** transcriptions.


## Conversion Details

| Property | Value |
|---|---|
| Source model | `MediaTek-Research/Breeze-ASR-26` |
| Architecture | Whisper Large V2 |
| Quantization | float16 |
| Model size | ~2.9 GB (vs ~6.2 GB float32 original) |
| CTranslate2 version | 4.7.1 |

## Usage

Run inference:

```python
from faster_whisper import WhisperModel

model = WhisperModel("MediaTek-Research/Breeze-ASR-26-ct2", device="cuda", compute_type="float16")
# For CPU: WhisperModel("...", device="cpu", compute_type="int8")

segments, info = model.transcribe("audio.wav", language="zh", task="transcribe")

for segment in segments:
    print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")
```

> **Note:** This model outputs **Mandarin Chinese characters** (not Taigi orthography / 台語正字). Pass `language="zh"` explicitly to avoid language detection overhead.


## Citation

```bibtex
@misc{lan2026breezetaigibenchmarksmodels,
      title={Breeze Taigi: Benchmarks and Models for Taiwanese Hokkien Speech Recognition and Synthesis},
      author={Yu-Siang Lan and Chia-Sheng Liu and Yi-Chang Chen and Po-Chun Hsu and Allyson Chiu and Shun-Wen Lin and Da-shan Shiu and Yuan-Fu Liao},
      year={2026},
      eprint={2603.19259},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2603.19259},
}
```

## License

Apache 2.0, same as the original model.