Pakorn2112/whisper-model-large-hmong

Whisper Large Hmong ASR

This repository provides an automatic speech recognition (ASR) model fine-tuned for the Hmong language using Whisper Large.

The model is available on Hugging Face and can be used immediately with the Transformers pipeline.

---

0. Model Description

This model is a fine-tuned version of Whisper Large for Hmong automatic speech recognition (ASR). It is trained on carefully prepared Hmong speech datasets ranging from short to medium-length utterances. The model is optimized for real-world conversational and general speech.

---

1. Quick Usage (Transformers Pipeline)

from transformers import pipeline

transcriber = pipeline(
    "automatic-speech-recognition",
    model="Pakorn2112/whisper-"
)

result = transcriber("hmong_sample.wav")
print(result["text"])

---

2. Dataset and Training Summary

Single Speech Model

• Datasets: 533 audio files
• Average audio duration: 1 to 20 seconds
• Author: Pakorn Archakeeree, localvoice.org
• Evaluation loss: 0.5448943376541138
• Word error rate (WER): 12.2924
• Maximum training steps: 10000

Multi Speech Model

• Datasets: 1017 audio files
• Average audio duration: 1 to 20 seconds
• Author: Pakorn Archakeeree, localvoice.org
• Evaluation loss: 0.6393303275108337
• Word error rate (WER): 17.6957
• Maximum training steps: 10000

---

3. Model Details

• Base model: Whisper Large
• Task: Automatic Speech Recognition (ASR)
• Target language: Hmong
• Framework: Hugging Face Transformers
• Format: safetensors

---

4. Acknowledgements

This work is supported by:

• HPC Ignite
• ThaiSC
• localvoice.org

Special thanks to the localvoice.org community for supporting Hmong language technology development.

---

5. License

This project is licensed under the Apache License 2.0. Please see the LICENSE file for details.