Fine tune whisper-large-v3-turbo-zh to enhance Chinese speech recognition capabilities, Belle-whisper-large-v3-turbo-zh demonstrates a 24-64% relative improvement in performance to whisper-large-v3-turbo on Chinese ASR benchmarks, including AISHELL1, AISHELL2, WENETSPEECH, and HKUST.

Same to Belle-whisper-large-v3-zh-punct, the punctuation marks come from model punc_ct-transformer_cn-en-common-vocab471067-large, and are added to the training datasets.

Usage

python

from transformers import pipeline

transcriber = pipeline(
  "automatic-speech-recognition", 
  model="BELLE-2/Belle-whisper-large-v3-turbo-zh"
)

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

transcription = transcriber("my_audio.wav") 

Fine-tuning

Model	(Re)Sample Rate	Train Datasets	Fine-tuning (full or peft)
Belle-whisper-large-v3-turbo-zh	16KHz	AISHELL-1 AISHELL-2 WenetSpeech HKUST	full fine-tuning

If you want to fine-thuning the model on your datasets, please reference to the github repo

CER(%) ↓

Model	Language Tag	aishell_1_test(↓)	aishell_2_test(↓)	wenetspeech_net(↓)	wenetspeech_meeting(↓)	HKUST_dev(↓)
whisper-large-v3	Chinese	8.085	5.475	11.72	20.15	28.597
whisper-large-v3-turbo	Chinese	8.639	6.014	13.507	20.313	37.324
Belle-whisper-large-v3-turbo-zh	Chinese	3.070	4.114	10.230	13.357	18.944

It is worth mentioning that compared to whisper-large-v3 and whisper-large-v3-turbo, Belle-whisper-large-v3-turbo-zh has a significant improvement.

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}},
}