Belle Distilwhisper Large V2 Zh

🚀 Belle-distilwhisper-large-v2-zh

This project fine-tunes distilwhisper-large-v2 to enhance Chinese speech recognition capabilities. It's 5.8 times faster and has 51% fewer parameters compared to whisper-large-v2, achieving a relative improvement of -3% to 35%.

🚀 Quick Start

If you find this model helpful, please like this model and star us on https://github.com/LianjiaTech/BELLE and https://github.com/shuaijiang/Whisper-Finetune

✨ Features

• High Efficiency: Similar to distilwhisper-large-v2, Belle-distilwhisper-large-v2-zh is 5.8 times faster and has 51% fewer parameters compared to whisper-large-v2.
• Improved Performance: Despite having 51% fewer parameters, it achieves a relative improvement of -3% to 35% over whisper-large-v2.
• Chinese Support: The original distilwhisper-large-v2 cannot transcribe Chinese (it only outputs English), while this model can handle Chinese speech recognition.

💻 Usage Examples

Basic Usage

from transformers import pipeline

transcriber = pipeline(
  "automatic-speech-recognition", 
  model="BELLE-2/Belle-distilwhisper-large-v2-zh"
)

transcriber.model.config.forced_decoder_ids = (
  transcriber.tokenizer.get_decoder_prompt_ids(
    language="zh", 
    task="transcribe"
  )
)

transcription = transcriber("my_audio.wav") 

📚 Documentation

Fine-tuning

If you want to fine-tune the model on your datasets, please refer to the github repo

CER(%) ↓

📄 License

This project is licensed under the Apache-2.0 license.

📚 Citation

Please cite our paper and github when using our code, data or model.

@misc{BELLE,
  author = {BELLEGroup},
  title = {BELLE: Be Everyone's Large Language model Engine},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/LianjiaTech/BELLE}},
}