predict_user_file.py

import os
import sys
import numpy as np
import librosa
import tensorflow as tf
from tensorflow.keras.models import load_model

# Ensure we can import utils
sys.path.append(os.getcwd())
try:
    from utils.hear_extractor import HeARExtractor
    from utils.audio_preprocessor import advanced_preprocess
except ImportError:
    sys.path.append(os.path.dirname(os.getcwd()))
    from utils.hear_extractor import HeARExtractor
    from utils.audio_preprocessor import advanced_preprocess

# Configuration
MODEL_PATH = r"c:\Users\ASUS\lung_ai_project\models\hear_classifier_advanced.h5"
CLASSES_PATH = r"c:\Users\ASUS\lung_ai_project\models\hear_classes_advanced.npy"
USER_FILE = r"C:\Users\ASUS\Downloads\WhatsApp Audio 2026-02-23 at 6.09.14 PM.wav"

def predict_single_file(file_path):
    print(f"\n--- Analyzing Audio: {os.path.basename(file_path)} ---")
    
    if not os.path.exists(file_path):
        print(f"Error: File not found at {file_path}")
        return

    # 1. Initialize Extractor
    print("Initializing HeAR Extractor...")
    try:
        extractor = HeARExtractor()
    except Exception as e:
        print(f"Failed to load HeAR model: {e}")
        return

    # 2. Load Evaluation Model
    print(f"Loading Model from {MODEL_PATH}...")
    try:
        model = load_model(MODEL_PATH, compile=False)
        classes = np.load(CLASSES_PATH)
        print(f"Classes: {classes}")
    except Exception as e:
        print(f"Error loading model: {e}")
        return

    # 3. Process & Predict
    try:
        # Load Audio
        print("Loading and preprocessing audio...")
        y, sr = librosa.load(file_path, sr=16000, duration=5.0)
        
        # --- NEW: Audio Validation (Gatekeeper) ---
        from utils.audio_validator import validate_audio_is_cough
        is_valid_cough, reason, val_conf = validate_audio_is_cough(y, sr)
        
        if not is_valid_cough:
            print("\n" + "="*50)
            print(f"REJECTED: Audio Validation Failed!")
            print(f"REASON: {reason}")
            print(f"RECOMMENDATION: Please record a clear, loud cough in a quiet room.")
            print("="*50)
            return

        # Apply Advanced Preprocessing (Critical for correct result!)
        y_clean = advanced_preprocess(y, sr)
        
        # Extract Embedding
        print("Extracting features...")
        emb = extractor.extract(y_clean)
        
        if emb is not None:
            # 4. Predict
            print("Step 4: Running Inference...")
            try:
                X = emb[np.newaxis, ...]
                preds = model.predict(X, verbose=0)
                pred_idx = np.argmax(preds[0])
                raw_label = classes[pred_idx]
                confidence = preds[0][pred_idx]
                
                # --- Reliability Guard ---
                THRESHOLD = 0.70
                if raw_label == "sick" and confidence < THRESHOLD:
                    print(f"DEBUG: Borderline result ({confidence:.2f}). Applying reliability guard.")
                    final_label = "healthy"
                    is_inconclusive = True
                else:
                    final_label = raw_label
                    is_inconclusive = False
                    
            except Exception as e:
                print(f"Error during inference: {e}")
                return

            print("\n" + "="*50)
            if is_inconclusive:
                print(f"RESULT: HEALTHY (Normal Pattern)")
                print(f"NOTE: Prediction was borderline ({confidence*100:.1f}%).")
                print("Reliability guard applied: No strong abnormal indicators found.")
            else:
                print(f"RESULT: {final_label.upper()}")
                print(f"CONFIDENCE: {confidence*100:.2f}%")
            print("="*50)
            
            # Simple interpretation
            if final_label == "sick":
                print("Recommendation: Potential respiratory symptoms detected. Consider medical consultation.")
            else:
                if is_inconclusive:
                    print("Recommendation: Recording had minor artifacts but appears normal. Re-record in a quiet room for better accuracy.")
                else:
                    print("Recommendation: Acoustic pattern appears healthy. Continue monitoring if symptoms persist.")
                
        else:
            print("Error: Could not extract features from audio.")
            
    except Exception as e:
        print(f"Detailed Error: {e}")

if __name__ == "__main__":
    predict_single_file(USER_FILE)