Whisper-Medium Arabic (Stage 2 Fine-Tuned FP16)

Fine-tuned Whisper-Medium trained with a two-stage curriculum strategy on Modern Standard Arabic, achieving 18.16% WER on Common Voice 17.0.

• High-throughput inference: For 8.8x realtime deployment, see the INT8 quantized version: Omar10lfc/whisper-medium-arabic-stage2-ct2-int8.

Arabic Speech Recognition Benchmark (Common Voice 17.0)

All models were evaluated on the canonical Mozilla Common Voice 17.0 Arabic test set using the standardized normalization protocol:

Model Details

• Base Architecture: openai/whisper-medium (769M parameters, 24 encoder layers, 24 decoder layers)
• Training Strategy: Two-Stage QLoRA Curriculum (4,000 steps total).
• Language: Arabic (ar)

FP16 (Unquantized) vs. CTranslate2 INT8 Quantization Comparison

Does 8-bit quantization degrade accuracy? Our empirical benchmark across the full Mozilla Common Voice Arabic evaluation suite demonstrates quantization parity—the Word Error Rate (WER) difference between unquantized FP16 and CTranslate2 Static INT8 is statistically negligible ($\le 0.5$ to $0.8$ percentage points), while reducing disk size and VRAM by ~50% and accelerating inference by up to $3.6\times$:

Production Recommendations

• Deploy CTranslate2 INT8 (*-ct2-int8): Recommended for production APIs, serverless containers (Modal, RunPod, Hugging Face Spaces), and edge environments to cut cold-start load times and memory by half while maximizing throughput.
• Deploy PyTorch FP16: Recommended for further domain fine-tuning, research workflows, or systems requiring native Hugging Face transformers pipeline integration.

Quickstart (Transformers)

import torch
from transformers import AutoModelForConditionalGeneration, AutoProcessor, pipeline

model_id = "Omar10lfc/whisper-medium-arabic-stage2"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForConditionalGeneration.from_pretrained(
    model_id, 
    torch_dtype=torch.float16, 
    device_map="cuda"
)

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    generate_kwargs={"language": "arabic", "task": "transcribe"}
)

result = pipe("path/to/arabic_audio.mp3")
print("Transcription:", result["text"])