Pakorn2112/whisper-model-large-hmong-multi-speech

Hmong Automatic Speech Recognition with Whisper

This project uses a fine-tuned Whisper model for Hmong Automatic Speech Recognition (ASR).

Model: Pakorn2112/whisper-model-large-hmong-multi-speech

The model is based on OpenAI Whisper and fine-tuned for recognizing Hmong speech more accurately.

About Hmong Language

Hmong is a language spoken by Hmong communities across several countries, including:

• —Thailand
• —Laos
• —Vietnam
• —China
• —United States

Hmong belongs to the Hmong-Mien language family and has multiple dialects, such as:

• —White Hmong (Hmoob Dawb)
• —Green/Blue Hmong (Moob Leeg)

Because Hmong is considered a low-resource language in speech technology, pretrained ASR systems often perform poorly without fine-tuning. This model helps improve speech recognition performance specifically for Hmong speech.

Model Information

• —Base Model: Whisper Large
• —Task: Automatic Speech Recognition (ASR)
• —Language: Hmong
• —Framework: Hugging Face Transformers
• —Model Hub: Pakorn2112/whisper-model-large-hmong-multi-speech

Installation

Install the required libraries:

pip install transformers datasets torchaudio librosa accelerate
pip install torch

Optional (recommended for audio processing):

pip install soundfile

Basic Usage

Load Model and Transcribe Audio

from transformers import pipeline

model_id = "Pakorn2112/whisper-model-large-hmong-multi-speech"

pipe = pipeline(
    "automatic-speech-recognition",
    model=model_id
)

result = pipe("your_audio.wav")

print(result["text"])

Better Configuration Example

For improved performance:

from transformers import pipeline
import torch

model_id = "Pakorn2112/whisper-model-large-hmong-multi-speech"

device = 0 if torch.cuda.is_available() else -1

pipe = pipeline(
    "automatic-speech-recognition",
    model=model_id,
    device=device
)

result = pipe(
    "your_audio.wav",
    generate_kwargs={
        "language": "hmong",
        "task": "transcribe"
    }
)

print(result["text"])

Using Processor + Model Directly

Advanced users can use AutoProcessor and AutoModelForSpeechSeq2Seq.

from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq
import torch

model_id = "Pakorn2112/whisper-model-large-hmong-multi-speech"

processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id)

model.eval()

print("Model loaded successfully")

Example Input

Audio:

hmong_speech.wav

Expected Output:

Kuv mus tom khw hnub no

English meaning:

I went to the market today

Notes

• —Recommended audio format: WAV
• —Recommended sample rate: 16kHz
• —Mono channel preferred
• —Cleaner audio gives better transcription quality

Citation

If you use this model in research or production, please cite:

@misc{pakorn_hmong_whisper,
  author = {Pakorn},
  title = {Whisper Large Fine-tuned for Hmong ASR},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Pakorn2112/whisper-model-large-hmong-multi-speech}}
}

Hugging Face Model Link

Model page:

https://huggingface.co/Pakorn2112/whisper-model-large-hmong-multi-speech

License

Apache-2.0