Luigi/whisper-small-zh_tw-ct2

Whisper Small (Fine-tuned for zh-TW) - CTranslate2 Format

This model is a CTranslate2-converted version of Jingmiao/whisper-small-zh_tw, which is a fine-tuned variant of OpenAI's whisper-small model, optimized for Taiwanese Mandarin (zh-TW) automatic speech recognition (ASR).

The conversion was done using the ct2-transformers-converter tool from CTranslate2, allowing for fast CPU inference with faster-whisper.

🔧 Usage (Python)

from faster_whisper import WhisperModel

model = WhisperModel("Luigi/whisper-small-zh_tw-ct2", device="cpu", compute_type="int8")

segments, _ = model.transcribe("your_audio_file.wav", beam_size=5)
transcription = "".join([seg.text for seg in segments])
print(transcription)

🗣️ Language

• Taiwanese Mandarin (zh-TW)

⚙️ Inference Backend

• Library: faster-whisper
• Format: CTranslate2
• Device: Optimized for CPU (int8 quantized)

📦 Files Included

• model.bin: The converted quantized model.
• vocab.json, tokenizer.json, config.json: Tokenizer and model configs.
• All files are compatible with the faster-whisper runtime.

💡 How to Convert Your Own Whisper Model

ct2-transformers-converter --model YOUR_MODEL_DIR \
  --output_dir YOUR_OUTPUT_DIR \
  --quantization int8 \
  --copy_files tokenizer.json

🧠 Credits

• Fine-tuned model by Jingmiao
• Conversion and deployment by Luigi
• Based on OpenAI Whisper
• CTranslate2 support from OpenNMT

🪪 License

Apache 2.0