Rabe3/saudi_stt

Whisper Large V3 Turbo - Saudi Arabic + Code-Switching

Fine-tuned version of openai/whisper-large-v3-turbo for Saudi Arabic dialect and Arabic-English code-switching.

Training Data

• Rabe3/SAD22_Cleaned (102k Saudi Arabic samples)
• MohamedRashad/arabic-english-code-switching (12k samples)

Training Details

• Base model: openai/whisper-large-v3-turbo
• Fine-tuning framework: whisper-finetune
• Epochs: 3
• Learning rate: 1e-5
• Batch size: 16

Usage

import whisper
import torch
from safetensors.torch import load_file
from huggingface_hub import hf_hub_download

# Download and load model
path = hf_hub_download(repo_id="Rabe3/saudi_stt", filename="model.safetensors")
model = whisper.load_model("large-v3-turbo", device="cuda")
state_dict = load_file(path, device="cuda")
model.load_state_dict(state_dict)
model.eval()

# Transcribe
result = model.transcribe(
    "audio.wav",
    language="ar",
    fp16=True,
    beam_size=5,
    best_of=5,
    temperature=0.0
)
print(result["text"])