CWTchen/Belle-whisper-large-v3-zh-punct-ct2-float32

Belle Faster Whisper Large V3 - Chinese with Punctuation

This is the CTranslate2 version of BELLE-2/Belle-whisper-large-v3-zh-punct, optimized for fast inference with faster-whisper.

Model Description

This model is a Chinese-optimized version of Whisper Large V3, fine-tuned specifically for Chinese speech recognition and punctuation prediction. By converting to the CTranslate2 format, it achieves faster inference speeds and lower memory usage.

Usage

Install Dependencies

pip install faster-whisper

Basic Usage

from faster_whisper import WhisperModel

# Load the model
model = WhisperModel("CWTchen/Belle-faster-whisper-large-v3-zh-punct", device="cuda", compute_type="float16")

# Transcribe an audio file
segments, info = model.transcribe("audio.mp3", language="zh")

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

Advanced Parameters

segments, info = model.transcribe(
    "audio.mp3",
    language="zh",
    beam_size=5,
    vad_filter=True,
    vad_parameters=dict(min_silence_duration_ms=500)
)

Examples

Here are some examples of transcription using this model. The audio files can be found in this GitHub repository.

Model Specifications

• Base Model: BELLE-2/Belle-whisper-large-v3-zh-punct
• Format: CTranslate2
• Quantization: float32
• Language: Traditional Chinese, Simplified Chinese
• Features: Automatic Punctuation

Conversion Command

This model was converted using the following command:

ct2-transformers-converter \
  --model BELLE-2/Belle-whisper-large-v3-zh-punct \
  --output_dir Belle-faster-whisper-large-v3-zh-punct \
  --copy_files tokenizer.json preprocessor_config.json \
  --quantization float32

Performance

Compared to the original Transformers version, the CTranslate2 version offers:

• Faster inference speed (approx. 2-4x)
• Lower memory usage
• Support for batch processing
• Support for VAD (Voice Activity Detection)

Citation

If you use this model, please cite the original BELLE model:

@misc{belle-whisper-v3-zh,
  author = {BELLE Team},
  title = {Belle Whisper Large V3 Chinese Punctuation},
  year = {2024},
  publisher = {Hugging Face},
  howpublished = {\url{[https://huggingface.co/BELLE-2/Belle-whisper-large-v3-zh-punct](https://huggingface.co/BELLE-2/Belle-whisper-large-v3-zh-punct)}}
}

License

Please refer to the license of the original model.

Acknowledgements

• Original model trained by the BELLE Team
• Based on OpenAI Whisper Large V3
• Converted using CTranslate2