app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import torch.nn.functional as F
import re
import string

# 1. Model Repository Configuration
MODEL_REPO = "SuperSl6/saudi-eou-model-v1"

print(f"Loading Model from {MODEL_REPO}...")
try:
    tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO)
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_REPO)
    model.eval()
    print("Model loaded successfully!")
except Exception as e:
    print(f"Error loading model: {e}")

# 2. Text Normalization Function
# This matches the preprocessing steps used during training to ensure accuracy.
def normalize_text(text):
    text = str(text)
    # Remove Arabic Diacritics (Tashkeel)
    text = re.sub(r'[\u0617-\u061A\u064B-\u0652]', '', text)
    # Normalize Alef forms (unify to bare Alef)
    text = re.sub(r'[أإآ]', 'ا', text)
    # Normalize Ya forms
    text = re.sub(r'ى', 'ي', text)
    # Remove Punctuation
    translator = str.maketrans('', '', string.punctuation + '،؛؟')
    text = text.translate(translator)
    return text.strip()

# 3. Prediction Function
def predict_eou(text):
    if not text or not text.strip():
        return "Please enter text...", "0%"
        
    # Clean text before inference
    clean_text = normalize_text(text)
    
    inputs = tokenizer(clean_text, return_tensors="pt", truncation=True, padding=True)
    
    with torch.no_grad():
        logits = model(**inputs).logits
        probs = F.softmax(logits, dim=1).numpy()[0]
    
    # Label 1 represents COMPLETE (EOU)
    score_complete = probs[1]
    
    # Define Confidence Threshold
    threshold = 0.75
    
    if score_complete >= threshold:
        decision = "REPLY (Turn Complete)"
    else:
        decision = "WAIT (Turn Incomplete)"
        
    # Return formatted decision and confidence percentage
    return decision, f"{score_complete:.1%}"

# 4. Gradio Interface Setup
examples = [
    ["السلام عليكم"],
    ["ياخي ودي أحجز"],
    ["ياخي ودي أحجز موعد عندكم بكرة"],
    ["رقم جوالي صفر خمسة"]
]

iface = gr.Interface(
    fn=predict_eou,
    inputs=gr.Textbox(label="User Speech Input", placeholder="Type here..."),
    outputs=[
        gr.Textbox(label="Model Decision"),
        gr.Label(label="Confidence Score")
    ],
    title="Saudi Dialect End-of-Utterance (EOU) Detector",
    description="A fine-tuned SaudiBERT model designed to detect end-of-utterance in real-time Saudi dialect conversations for AI voice agents.",
    examples=examples
)

iface.launch()