Arabic End-of-Utterance (EOU) Detection Model

Model Description

Fine-tuned model for Arabic End-of-Utterance detection, optimized for Saudi dialect conversations. Designed for real-time integration with LiveKit voice agents.

Performance Metrics (Step 2400)

Metric Value
F1 Score 0.534
Precision 0.431
Recall 0.702
FPR 0.150

Intended Use

  • Real-time voice agent turn detection
  • Arabic conversational AI systems
  • Saudi dialect speech processing

Training Details

  • Base Model: [specify your base model]
  • Training Steps: 2400
  • Validation Loss: 0.462
  • Training Date: December 2024

Usage

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained("{username}/{repo_name}")
tokenizer = AutoTokenizer.from_pretrained("{username}/{repo_name}")

# Example inference
text = "نعم، أنا أفهم ما تقصد"
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
eou_probability = torch.softmax(outputs.logits, dim=-1)[0][1].item()

Limitations

  • Optimized for Saudi dialect
  • May require threshold tuning for specific use cases
  • Designed for conversational contexts

Citation

@misc{arabic-eou-2024,
  author = {Your Name},
  title = {Arabic EOU Detection Model},
  year = {2024},
  publisher = {HuggingFace},
  url = {https://huggingface.co/{username}/{repo_name}}
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support