Model Card for distill-whisper-th-large-v3-ct2

distill-whisper-th-large-v3-ct2 is a CTranslate2-converted version of biodatlab/distill-whisper-th-large-v3, a distilled and Thai-optimized Whisper model for automatic speech recognition (ASR). This model is compatible with faster-whisper, allowing for faster and more efficient transcription, especially on low-resource hardware.

Model Details

Model Description

• Developed by: BioDataLab
• Model type: CTranslate2
• Languages: Thai (th)
• Finetuned from: openai/whisper-large-v3-turbo → distilled and specialized for Thai
• Quantization: float16

Model Sources

• Original model: https://huggingface.co/biodatlab/distill-whisper-th-large-v3
• CTranslate2 project: https://github.com/OpenNMT/CTranslate2

Conversion command:

ct2-transformers-converter \
  --model biodatlab/distill-whisper-th-large-v3 \
  --output_dir distill-whisper-th-large-v3-ct2 \
  --copy_files tokenizer.json \
  --quantization float16