raghad23/arabic_eou_model

Arabic EOU Model – Saudi Dialect (raghad23/arabic-eou-model90)

High-performance End-of-Utterance (EOU) detection model for real-time Arabic voice agents (LiveKit, etc.).

Task

Binary classification:

• LABEL_0 → Speaker continues
• LABEL_1 → End of turn (EOU)

Performance (Test Set)

Metric	Score
Accuracy	90.68%
Weighted F1	93.40%
Precision (EOU)	99.83%
Recall (EOU)	87.75%
Eval Loss	0.1964

Excellent real-time performance — detects end-of-utterance accurately in Saudi Arabic dialect.

Dataset

414k Saudi dialect samples
https://huggingface.co/datasets/raghad23/arabic_eou_sada_dataset

Base Model

aubmindlab/bert-base-arabertv02

Latency

< 80ms → perfect for LiveKit deployment

Usage

from transformers import pipeline
pipe = pipeline("text-classification", model="raghad23/arabic-eou-model90")
pipe("تمام الحمدلله وشلونك")  # → LABEL_1 = EOU