---
library_name: transformers
license: cc-by-nc-4.0
language:
- en
- zh
metrics:
- cer
pipeline_tag: automatic-speech-recognition
base_model:
- openai/whisper-large-v3-turbo
---

[ [繁體中文 README.md](https://huggingface.co/NUTN-KWS/Whisper-Taiwanese-model-v0.5) ]

# 👳 Whisper-Taiwanese model V0.5 (Tv0.5)

This model is a fine-tuned version of OpenAI’s [openai/whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo). It was developed by the National University of Tainan (NUTN), Taiwan, as part of a National Science and Technology Council (NSTC)-funded industry-academia collaboration project. We carried out the Taiwanese-English Co-Learning Pilot Project from September 2024 to June 2025 in collaboration with JEN-PIN ENTERPRISE CO., LTD. The model is trained for Taiwanese language recognition tasks using JEN-PIN educational materials generated through Student–Machine Co-Learning during the Fall 2024 semester. Additionally, the NUTN is collaborating with the National Center for High-performance Computing (NCHC) of the National Applied Research Laboratories (NARLabs) in Taiwan to provide computational and storage resources and co-develop an AI learning model for elementary and high school students.

Demo: [https://kws.oaselab.org/taigitong/](https://kws.oaselab.org/taigitong/)

## 📝 Model Details
- **Base Model**: `openai/whisper-large-v3-turbo`
- **Fine-tuned for**: Taiwanese Hokkien Automatic Speech Recognition (ASR)
- **Fine-tuning Framework**: Hugging Face Transformers
- **Training Duration**: Approximately 180 hours using two V100 GPUs
- **Dataset**: Custom dataset, including the Dictionary of Frequently-Used Taiwanese Taigi released by the Ministry of Education, Taiwan, totaling approximately 90 hours of audio data.
- **Input Format**: 16kHz mono WAV
- **License**: CC BY-NC 4.0

## 🚀 Usage
### Installing Packages:
```bash
pip install torch torchvision torchaudio transformers
```
### Example:
```python
from transformers import pipeline

pipe = pipeline("automatic-speech-recognition", model="./model/whisper-taiwanese", device=0)
result = pipe("audio.wav", generate_kwargs={"language": "zh", "task": "transcribe"})
print(result["text"])
```

## 👨‍🎓 Citation

### BibTeX:
```bibtex
@misc{taiwanesewhisperasr2025,
  title={Taiwanese Whisper ASR},
  author={KWS Center, National University of Tainan, Taiwan},
  year={2025},
  url={https://huggingface.co/NUTN-KWS/Whisper-Taiwanese-model-v0.5}
}
```

### APA:
- C. S. Lee, M. H. Wang, C. C. Yue, G. Y. Teseng, and Y. Nojima, "Fuzzy Estimation Agent with Knowledge Graph and Quantum Fuzzy Inference Engine for Taiwanese-English Co-Learning," 2025 IFSA World Congress and NAFIPS Annual Meeting (IFSA/NAFIPS 2025), Banff, Alberta, Canada, Aug. 16-19, 2025.
- C. S. Lee, M. H. Wang, C. Y. Chen, S. C. Yang, M. Reformat, N. Kubota, and A. Pourabdollah, "Integrating quantum CI and generative AI for Taiwanese/English co-learning," Quantum Machine Intelligence, vol. 6, 64, pp. 1-19, 2024.
- C. S. Lee, M. H. Wang, C. Y. Chen, S. C. Yang, M. Reformat, N. Kubota, and A. Pourabdollah, "Quantum fuzzy inference engine with generative AI and TAIDE KG for Taiwanese/English co-learning," 2025 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE 2025), Reims, France, Jul. 6-9, 2025.