Dragneel/whisper-medium-nepali-openslr

Whisper Medium Fine-tuned on Nepali (OpenSLR 54)

This model is a fine-tuned version of openai/whisper-medium on the OpenSLR 54 (Nepali Speech Corpus) dataset.

Model Details

• Model: Whisper Medium (769M Parameters)
• Dataset: ~154 Hours of Nepali Audio
• Language: Nepali
• Hardware: NVIDIA A100 80GB

Results

• WER: 14.04%
• Loss: 0.0710

Usage

from transformers import pipeline

transcriber = pipeline(
    "automatic-speech-recognition", 
    model="Dragneel/whisper-medium-nepali-openslr"
)

text = transcriber("path_to_audio.mp3")
print(text["text"])