Update app.py

app.py

@@ -719,7 +719,7 @@ def count_precise_apneas(y, sr, segment_results, min_apnea=1, max_apnea=60):
     # Check B: ML model score (gap must score < 0.40)
     # BOTH must pass. This fixes the padding problem and reduces false positives.
     print("   Stage 3: DUAL validation (RMS energy + ML score)...")
-    print("      Rules: gap RMS ratio < 0.5 AND ML score < 0.40")
+    print("      Rules: gap RMS ratio < 0.55 AND ML score < 0.40")
 
     validated_apneas = []
 
@@ -752,7 +752,7 @@ def count_precise_apneas(y, sr, segment_results, min_apnea=1, max_apnea=60):
             # Avoid division by zero
             rms_ratio = gap_rms / (avg_snore_rms + 1e-10)
 
-            rms_ok = rms_ratio < 0.5  # Gap must be at least 2x quieter than snoring
+            rms_ok = rms_ratio < 0.55  # Relaxed from 0.5 to 0.55 to account for noise reduction
 
         except Exception as e:
             print(f"      ⚠️ RMS check error for candidate {idx+1}: {e}")
@@ -834,7 +834,7 @@ def count_precise_apneas(y, sr, segment_results, min_apnea=1, max_apnea=60):
         else:
             reasons = []
             if not rms_ok:
-                reasons.append(f"gap not quiet enough (RMS ratio {rms_ratio:.3f} >= 0.5)")
+                reasons.append(f"gap not quiet enough (RMS ratio {rms_ratio:.3f} >= 0.55)")
             if not ml_ok:
                 reasons.append(f"ML says snoring in gap (score {ml_score:.3f} >= 0.40)")
             print(f"         → REJECTED: {', '.join(reasons)}")
@@ -847,7 +847,7 @@ def count_precise_apneas(y, sr, segment_results, min_apnea=1, max_apnea=60):
 # VALIDATION
 # ========================================
 def validate_sleep_recording(y, sr):
-    """Validate that this is a sleep recording"""
+    """Validate that this is a sleep recording with detailed diagnostics"""
     duration = len(y) / sr
     
     if duration < 20:
@@ -855,12 +855,11 @@ def validate_sleep_recording(y, sr):
     
     rms_energy = np.sqrt(np.mean(y**2))
     
-    if rms_energy < 0.001:
-        return False, 0.0, "Audio is blank/silent"
-    
-    if rms_energy > 0.6:
-        return False, 0.0, "Audio too loud (not a sleep recording)"
+    # Calculate zero-crossing rate
+    zcr = librosa.feature.zero_crossing_rate(y)[0]
+    zcr_mean = np.mean(zcr)
     
+    # Calculate spectral characteristics
     stft = np.abs(librosa.stft(y))
     freq_bins = librosa.fft_frequencies(sr=sr)
     
@@ -871,11 +870,27 @@ def validate_sleep_recording(y, sr):
     low_freq_ratio = low_freq_energy / total_energy
     high_freq_ratio = high_freq_energy / total_energy
     
+    # DIAGNOSTIC LOGGING
+    print(f"   📊 AUDIO QUALITY DIAGNOSTICS:")
+    print(f"      RMS Energy: {rms_energy:.4f} (normal: 0.001-0.6)")
+    print(f"      Zero-Crossing Rate: {zcr_mean:.4f} (speech-like if > 0.2)")
+    print(f"      Low Freq (<2kHz): {low_freq_ratio:.3f} (should be > 0.4)")
+    print(f"      High Freq (>4kHz): {high_freq_ratio:.3f} (should be < 0.3)")
+    
+    if rms_energy < 0.001:
+        return False, 0.0, "Audio is blank/silent"
+    
+    if rms_energy > 0.6:
+        return False, 0.0, "Audio too loud (not a sleep recording)"
+    
     if low_freq_ratio < 0.4:
-        return False, 0.0, "Frequency distribution doesn't match sleep audio"
+        return False, 0.0, f"Frequency distribution doesn't match sleep audio (low_freq={low_freq_ratio:.3f}, needs >0.4) - FRONTEND AUDIO QUALITY ISSUE"
     
     if high_freq_ratio > 0.3:
-        return False, 0.0, "Too much high-frequency content (speech/TV detected)"
+        return False, 0.0, f"Too much high-frequency content (high_freq={high_freq_ratio:.3f}, needs <0.3) - FRONTEND RECORDING HAS NOISE/ARTIFACTS"
+    
+    if zcr_mean > 0.2:
+        return False, 0.0, f"Zero-crossing rate too high ({zcr_mean:.3f}) - Audio sounds like speech/noise, not snoring - FRONTEND RECORDING ISSUE"
     
     return True, 1.0, "Valid sleep recording detected"