Update app.py

app.py

@@ -5,6 +5,7 @@ import torch
 from transformers import Wav2Vec2Processor, Wav2Vec2Model
 from simple_salesforce import Salesforce
 import os
+import hashlib
 from datetime import datetime
 
 # Salesforce credentials (store securely in environment variables)
@@ -29,30 +30,40 @@ except Exception as e:
 processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
 model = Wav2Vec2Model.from_pretrained("facebook/wav2vec2-base-960h")
 
+def compute_file_hash(file_path):
+    """Compute MD5 hash of a file to check uniqueness."""
+    hash_md5 = hashlib.md5()
+    with open(file_path, "rb") as f:
+        for chunk in iter(lambda: f.read(4096), b""):
+            hash_md5.update(chunk)
+    return hash_md5.hexdigest()
+
 def analyze_voice(audio_file):
     """Analyze voice for health indicators."""
     try:
         # Log audio file info
-        print(f"Processing audio file: {audio_file}")
+        file_hash = compute_file_hash(audio_file)
+        print(f"Processing audio file: {audio_file}, Hash: {file_hash}")
         
         # Load audio file
         audio, sr = librosa.load(audio_file, sr=16000)
-        print(f"Audio shape: {audio.shape}, Sampling rate: {sr}, Duration: {len(audio)/sr:.2f}s")
+        audio = audio / (np.max(np.abs(audio)) + 1e-10)  # Normalize audio
+        print(f"Audio shape: {audio.shape}, Sampling rate: {sr}, Duration: {len(audio)/sr:.2f}s, Mean: {np.mean(audio):.4f}, Std: {np.std(audio):.4f}")
         
         # Process audio for Wav2Vec2
         inputs = processor(audio, sampling_rate=16000, return_tensors="pt", padding=True)
-        print(f"Input tensor shape: {inputs['input_values'].shape}")
+        print(f"Input tensor shape: {inputs['input_values'].shape}, Sample values: {inputs['input_values'][0][:5]}")
         
         with torch.no_grad():
             outputs = model(**inputs)
         
         # Extract features
-        features = outputs.last_hidden_state.mean(dim=1).numpy()
-        print(f"Features shape: {features.shape}, Sample values: {features[0][:5]}")
+        features = outputs.last_hidden_state.numpy()  # Use full hidden states
+        print(f"Features shape: {features.shape}, Sample values: {features[0, 0, :5]}")
         
-        # Mock health analysis
-        respiratory_score = np.mean(features)
-        mental_health_score = np.std(features)
+        # Compute scores
+        respiratory_score = np.mean(features, axis=(1, 2))
+        mental_health_score = np.std(features, axis=(1, 2))
         
         # Log scores
         print(f"Respiratory Score: {respiratory_score:.4f}, Mental Health Score: {mental_health_score:.4f}")
@@ -61,13 +72,13 @@ def analyze_voice(audio_file):
         feedback = ""
         if respiratory_score > 0.1:
             feedback += f"Possible respiratory issue detected (score: {respiratory_score:.4f}); consult a doctor. "
-        if mental_health_score > 0.1:
+        if mental_health_score > 0.2:  # Raised from 0.1 to reduce false positives
             feedback += f"Possible stress indicators detected (score: {mental_health_score:.4f}); consider professional advice. "
         
         if not feedback:
             feedback = "No significant health indicators detected."
         
-        feedback += f"\n\n**Debug Info**: Respiratory Score = {respiratory_score:.4f}, Mental Health Score = {mental_health_score:.4f}"
+        feedback += f"\n\n**Debug Info**: Respiratory Score = {respiratory_score:.4f}, Mental Health Score = {mental_health_score:.4f}, File Hash = {file_hash}"
         feedback += "\n**Disclaimer**: This is not a diagnostic tool. Consult a healthcare provider for medical advice."
         
         # Store in Salesforce