Upload folder using huggingface_hub

app/main.py

@@ -1,158 +1,170 @@
-import os
-import sys
-
-# --- Force Writable Paths for Hugging Face ---
-# This MUST happen before any other imports
-os.environ['HOME'] = '/tmp'
-os.environ['HF_HOME'] = '/tmp/huggingface'
-os.environ['XDG_CACHE_HOME'] = '/tmp/cache'
-os.environ['MPLCONFIGDIR'] = '/tmp/matplotlib'
-os.environ['NUMBA_CACHE_DIR'] = '/tmp/numba'
-
-# Ensure directories exist
-for d in ['/tmp/huggingface', '/tmp/cache', '/tmp/matplotlib', '/tmp/numba']:
-    os.makedirs(d, exist_ok=True)
-
-import numpy as np
-import librosa
-import tensorflow as tf
-from flask import Flask, request, jsonify
-from flask_cors import CORS
-from tensorflow.keras.models import load_model
-from werkzeug.utils import secure_filename
-
-# --- Absolute Path Resolution ---
-# This ensures the Linux server finds the 'utils' and 'models' folders
-current_dir = os.path.dirname(os.path.abspath(__file__))
-project_root = os.path.dirname(current_dir)
-if project_root not in sys.path:
-    sys.path.insert(0, project_root)
-
-try:
-    from utils.hear_extractor import HeARExtractor
-    from utils.audio_preprocessor import advanced_preprocess
-    print("✅ Successfully imported utils package.")
-except ImportError as e:
-    print(f"❌ Critical Import Error: {e}")
-    # Show more context for debugging
-    print(f"DEBUG: sys.path is {sys.path}")
-    print(f"DEBUG: Attempting to look in {project_root}")
-    sys.exit(1)
-
-# Configuration
-BASE_DIR = os.path.dirname(os.path.abspath(__file__))
-app = Flask(__name__)
-# Enable CORS for the entire API to allow Vercel and Flutter to connect securely
-CORS(app)
-
-app.config['UPLOAD_FOLDER'] = os.path.join(BASE_DIR, 'tmp', 'uploads')
-app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024  # 16MB limit
-
-# Ensure upload directory exists
-os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
-
-# Configuration
-MODEL_PATH = os.path.join(project_root, "models", "hear_classifier_v9_ultimate.h5")
-
-# Global variables for lazy loading
-extractor = None
-classifier_model = None
-
-def load_resources():
-    global extractor, classifier_model
-    if extractor is None:
-        print("Initializing HeAR Extractor...")
-        hf_token = os.environ.get('HF_TOKEN')
-        extractor = HeARExtractor(token=hf_token)
-    
-    if classifier_model is None:
-        print(f"Loading Ultimate Model from {MODEL_PATH}...")
-        classifier_model = load_model(MODEL_PATH, compile=False)
-
-@app.route('/')
-def index():
-    return jsonify({
-        "status": "online",
-        "service": "KasaHealth Diagnostic API (Ultimate V9)",
-        "version": "1.3.0",
-        "message": "Send audio files via POST to /predict"
-    })
-
-@app.route('/predict', methods=['POST'])
-def predict():
-    if 'audio' not in request.files:
-        return jsonify({"error": "No audio file provided"}), 400
-    
-    file = request.files['audio']
-    if file.filename == '':
-        return jsonify({"error": "No selected file"}), 400
-    
-    if file:
-        filename = secure_filename(file.filename)
-        filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
-        file.save(filepath)
-        
-        try:
-            load_resources()
-            y, sr = librosa.load(filepath, sr=16000)
-            
-            # VAD Lite
-            rms_energy = np.mean(librosa.feature.rms(y=y))
-            if rms_energy < 0.005:
-                os.remove(filepath)
-                return jsonify({"error": "No cough detected. Please record in a quieter area."}), 400
-                
-            # 2. Preprocess to remove phone microphone rumble and static!
-            y_clean = advanced_preprocess(y, sr)
-            emb = extractor.extract(y_clean)
-            
-            if emb is not None:
-                X = emb[np.newaxis, ...]
-                prob = classifier_model.predict(X, verbose=0)[0][0]
-                
-                # Logic: Highly sensitive (Unleashed mode)
-                THRESHOLD = 0.50
-                
-                if prob > THRESHOLD:
-                    final_label = "sick"
-                    # If it's sick but barely (0.5 - 0.65), call it inconclusive
-                    is_inconclusive = bool(prob < 0.65)
-                    confidence = float(prob)
-                else:
-                    final_label = "healthy"
-                    # If it's healthy but close to sick, call it inconclusive
-                    is_inconclusive = bool(prob > 0.35)
-                    
-                    # Convert 0.0 -> 0.5 range to 1.0 -> 0.5 healthy confidence
-                    confidence = 1.0 - float(prob)
-                
-                raw_label = "sick" if prob > 0.5 else "healthy"
-                
-                os.remove(filepath)
-                return jsonify({
-                    "status": "success",
-                    "result": final_label,
-                    "confidence": confidence,
-                    "is_inconclusive": is_inconclusive,
-                    "raw_label": raw_label,
-                    "recommendation": get_recommendation(final_label, is_inconclusive)
-                })
-            else:
-                os.remove(filepath)
-                return jsonify({"error": "Feature extraction failed"}), 500
-                
-        except Exception as e:
-            if os.path.exists(filepath): os.remove(filepath)
-            return jsonify({"error": str(e)}), 500
-
-def get_recommendation(label, is_inconclusive):
-    if label == "sick":
-        return "Potential respiratory symptoms detected. We strongly recommend consulting a healthcare professional for a detailed evaluation."
-    elif is_inconclusive:
-        return "Acoustic signals show some variation but no strong abnormal indicators were found. Re-record in a quiet environment for more certainty."
-    else:
-        return "Acoustic pattern appears healthy. Continue to monitor your health and maintain good respiratory hygiene."
-
-if __name__ == '__main__':
-    # For local development
-    app.run(debug=True, port=5000)
+import os
+import sys
+
+# --- Force Writable Paths for Hugging Face ---
+os.environ['HOME'] = '/tmp'
+os.environ['HF_HOME'] = '/tmp/huggingface'
+os.environ['XDG_CACHE_HOME'] = '/tmp/cache'
+os.environ['MPLCONFIGDIR'] = '/tmp/matplotlib'
+os.environ['NUMBA_CACHE_DIR'] = '/tmp/numba'
+
+# Ensure directories exist
+for d in ['/tmp/huggingface', '/tmp/cache', '/tmp/matplotlib', '/tmp/numba']:
+    os.makedirs(d, exist_ok=True)
+
+import numpy as np
+import librosa
+import tensorflow as tf
+from flask import Flask, request, jsonify
+from flask_cors import CORS
+from tensorflow.keras.models import load_model
+from werkzeug.utils import secure_filename
+from scipy.signal import butter, lfilter
+
+# --- Absolute Path Resolution ---
+current_dir = os.path.dirname(os.path.abspath(__file__))
+project_root = os.path.dirname(current_dir)
+if project_root not in sys.path:
+    sys.path.insert(0, project_root)
+
+try:
+    from utils.hear_extractor import HeARExtractor
+    print("✅ Successfully imported utils package.")
+except ImportError as e:
+    print(f"❌ Critical Import Error: {e}")
+    sys.exit(1)
+
+app = Flask(__name__)
+CORS(app)
+
+app.config['UPLOAD_FOLDER'] = os.path.join('/tmp', 'uploads')
+app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024
+os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
+
+# Configuration: DUAL-BRAIN MODELS
+# Brain 1: The Shield (V9) - Protects healthy users
+# Brain 2: The Sentry (V10) - High sensitivity for sick patients
+MODEL_V9_PATH = os.path.join(project_root, "models", "hear_classifier_v9_ultimate.h5")
+MODEL_V10_PATH = os.path.join(project_root, "models", "hear_classifier_v10_sentry.h5")
+
+# Global variables for lazy loading
+extractor = None
+shield_model = None
+sentry_model = None
+
+def load_resources():
+    global extractor, shield_model, sentry_model
+    if extractor is None:
+        hf_token = os.environ.get('HF_TOKEN')
+        extractor = HeARExtractor(token=hf_token)
+    
+    if shield_model is None:
+        print(f"Loading Shield Model (V9)...")
+        shield_model = load_model(MODEL_V9_PATH, compile=False)
+    
+    if sentry_model is None:
+        print(f"Loading Sentry Model (V10)...")
+        sentry_model = load_model(MODEL_V10_PATH, compile=False)
+
+def highpass_filter(data, cutoff, fs, order=5):
+    nyq = 0.5 * fs
+    normal_cutoff = cutoff / nyq
+    b, a = butter(order, normal_cutoff, btype='high', analog=False)
+    return lfilter(b, a, data)
+
+@app.route('/')
+def index():
+    return jsonify({
+        "status": "online",
+        "service": "KasaHealth Dual-Brain Engine",
+        "version": "2.0.0 (Ensemble)",
+        "message": "Dual-Brain analysis ready."
+    })
+
+@app.route('/predict', methods=['POST'])
+def predict():
+    if 'audio' not in request.files:
+        return jsonify({"error": "No audio file"}), 400
+    
+    file = request.files['audio']
+    filename = secure_filename(file.filename)
+    filepath = os.path.join(app.config['UPLOAD_FOLDER'], filename)
+    file.save(filepath)
+    
+    try:
+        load_resources()
+        y, sr = librosa.load(filepath, sr=16000)
+        
+        # VAD Power check
+        rms_energy = np.mean(librosa.feature.rms(y=y))
+        if rms_energy < 0.005:
+            os.remove(filepath)
+            return jsonify({"error": "Silence detected. Please cough closer to the mic."}), 400
+            
+        # Pre-process
+        y_clean = highpass_filter(y, 100, sr) 
+        y_clean = y_clean / (np.max(np.abs(y_clean)) + 1e-8)
+        
+        emb = extractor.extract(y_clean)
+        if emb is None:
+            os.remove(filepath)
+            return jsonify({"error": "Failed to extract acoustic features."}), 500
+            
+        X = emb[np.newaxis, ...]
+        
+        # --- Dual-Brain Analysis ---
+        p9 = shield_model.predict(X, verbose=0)[0][0]
+        p10 = sentry_model.predict(X, verbose=0)[0][0]
+        
+        # Logic Mapping to Frontend Risk Meter:
+        # result: 'healthy' or 'sick'
+        # confidence: drives the meter (High/Med/Low)
+        
+        is_sick_sentry = bool(p10 > 0.50)
+        is_sick_shield = bool(p9 > 0.55) # Shield is stricter
+        
+        if not is_sick_sentry:
+            # Both agree it is likely fine
+            final_label = "healthy"
+            confidence = float(1.0 - p10) 
+            # Force high healthy confidence for Low Risk trigger
+            if confidence < 0.70: confidence = 0.75 
+            is_inconclusive = False
+            
+        elif is_sick_sentry and not is_sick_shield:
+            # Sentry catches something, but Shield isn't sure
+            final_label = "sick"
+            # In app.js: sick + conf < 0.75 = MEDIUM RISK
+            confidence = 0.60 
+            is_inconclusive = True
+            
+        else:
+            # Both models detect strong signs
+            final_label = "sick"
+            # In app.js: sick + conf >= 0.75 = HIGH RISK
+            confidence = 0.90
+            is_inconclusive = False
+
+        os.remove(filepath)
+        return jsonify({
+            "status": "success",
+            "result": final_label,
+            "confidence": confidence,
+            "is_inconclusive": is_inconclusive,
+            "scores": { "shield": float(p9), "sentry": float(p10) },
+            "recommendation": get_recommendation(final_label, is_inconclusive)
+        })
+            
+    except Exception as e:
+        if os.path.exists(filepath): os.remove(filepath)
+        return jsonify({"error": str(e)}), 500
+
+def get_recommendation(label, is_inconclusive):
+    if label == "sick":
+        if is_inconclusive:
+            return "Minor acoustic anomalies detected. We recommend re-testing in a quiet environment or monitoring symptoms."
+        return "Significant respiratory indicators found. We strongly recommend consulting a healthcare professional."
+    return "Clear acoustic signature. Maintain good respiratory hygiene."
+
+if __name__ == '__main__':
+    app.run(host='0.0.0.0', port=7860)

models/hear_classifier_v10_sentry.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1ec482f249e5ef8b6ebd970373c13014658f3e7c1e65c89014d5e6e5ac65206
+size 14787768

test_real_pipeline.py

@@ -19,7 +19,7 @@ from utils.audio_preprocessor import advanced_preprocess
 AUDIO_ROOT = r"c:\Users\ASUS\lung_ai_project\data\coughvid_public\organized"
 SICK_DIR = os.path.join(AUDIO_ROOT, "sick")
 HEALTHY_DIR = os.path.join(AUDIO_ROOT, "healthy")
-MODEL_PATH = os.path.join(PROJECT_ROOT, "models", "hear_classifier_v9_ultimate.h5")
+MODEL_PATH = os.path.join(PROJECT_ROOT, "models", "hear_classifier_v10_sentry.h5")
 
 META_PATH = r"c:\Users\ASUS\lung_ai_project\data\coughvid_public\metadata_compiled.csv"
 import pandas as pd
@@ -76,8 +76,16 @@ def test_pipeline():
                 # API rejects quiet audio. For testing, skip.
                 continue
                 
-            # 3. API Preprocess
-            y_clean = y
+            # 3. API Preprocess (Gentle Field Preprocess)
+            from scipy.signal import butter, lfilter
+            def highpass_filter(data, cutoff, fs, order=5):
+                nyq = 0.5 * fs
+                normal_cutoff = cutoff / nyq
+                b, a = butter(order, normal_cutoff, btype='high', analog=False)
+                return lfilter(b, a, data)
+            
+            y_clean = highpass_filter(y, 100, sr) 
+            y_clean = y_clean / (np.max(np.abs(y_clean)) + 1e-8)
             
             # 4. API Extract
             emb = extractor.extract(y_clean)