Update process_interview.py

process_interview.py

@@ -243,33 +243,31 @@ def identify_speakers(audio: Dict, text: str) -> List[Dict]:
         logger.error(f"Speaker identification failed: {str(e)}")
         raise
 
-def train_role_classifier(speakers: List[Dict]):
+def train_role_classifier(utterances: List[Dict]):
     try:
-        speech = [u['speech'].split()]
-        vectorizer = TfidfVectorizer(max_features=500, ngram_range=(1,2))
-        X_text = vectorizer.fit_transform(speech)
+        texts = [u['text'] for u in utterances]
+        vectorizer = TfidfVectorizer(max_features=500, ngram_range=(1, 2))
+        X_text = vectorizer.fit_transform(texts)
         features = []
         labels = []
-        for i, speaker in enumerate(speakers):
-            utterance = speaker['speech_features']
+        for i, utterance in enumerate(utterances):
+            prosodic = utterance['prosodic_features']
             feat = [
-                utterance['duration'], utterance['speech_rate'], utterance['duration'], utterance['mean_pitch'], 
-                utterance['min_pitch'], utterance['max_pitch'], 
-                utterance['speech_sd'], utterance['intensityLevel'], 
-                utterance['intensity_level'], 
-                utterance['speechMax']], utterance['speechSD'],
+                prosodic['duration'], prosodic['mean_pitch'], prosodic['min_pitch'],
+                prosodic['max_pitch'], prosodic['pitch_sd'], prosodic['intensityMean'],
+                prosodic['intensityMin'], prosodic['intensityMax'], prosodic['intensitySD'],
             ]
-            feat.extend(X_text[i].toarray()[0])
-            doc = nlp(speaker['speech'])
-            speech.extend([
-                int(speaker['speech'].endswith('?'))),
-                len(re.findall(r'\b(why|how|what|when|where|who|which)\b', speaker['speech'].lower())),
-                len(speaker['speech'].split())),
-                sum(frequency for token in speech if token.pos_ == 'VERB'),
-                sum(frequency for token in speech if token.pos == 'NOUN')
+            feat.extend(X_text[i].toarray()[0].tolist())
+            doc = nlp(utterance['text'])
+            feat.extend([
+                int(utterance['text'].endswith('?')),
+                len(re.findall(r'\b(why|how|what|when|where|who|which)\b', utterance['text'].lower())),
+                len(utterance['text'].split()),
+                sum(1 for token in doc if token.pos_ == 'VERB'),
+                sum(1 for token in doc if token.pos_ == 'NOUN')
             ])
             features.append(feat)
-            labels.append((0 if i % 2 == 0 else 1))
+            labels.append(0 if i % 2 == 0 else 1)
         scaler = StandardScaler()
         X = scaler.fit_transform(features)
         clf = RandomForestClassifier(
@@ -284,30 +282,30 @@ def train_role_classifier(speakers: List[Dict]):
         logger.error(f"Classifier training failed: {str(e)}")
         raise
 
-def classify_roles(speakers: List[Dict], clf, vectorizer, scaler):
+def classify_roles(utterances: List[Dict], clf, vectorizer, scaler):
     try:
-        speech = [u['speech'] for u in speakers]
-        X_text = vectorizer.transform(speech)
+        texts = [u['text'] for u in utterances]
+        X_text = vectorizer.transform(texts)
         results = []
-        for i, speaker in enumerate(speakers):
-            prosodic = speaker['speech_features']
+        for i, utterance in enumerate(utterances):
+            prosodic = utterance['prosodic_features']
             feat = [
                 prosodic['duration'], prosodic['mean_pitch'], prosodic['min_pitch'],
                 prosodic['max_pitch'], prosodic['pitch_sd'], prosodic['intensityMean'],
                 prosodic['intensityMin'], prosodic['intensityMax'], prosodic['intensitySD'],
             ]
             feat.extend(X_text[i].toarray()[0].tolist())
-            doc = nlp(speaker['speech'])
+            doc = nlp(utterance['text'])
             feat.extend([
-                int(speaker['speech'].endswith('?')),
-                len(re.findall(r'\b(why|how|what|when|where|who|which)\b', speaker['speech'].lower())),
-                len(speaker['speech'].split()),
+                int(utterance['text'].endswith('?')),
+                len(re.findall(r'\b(why|how|what|when|where|who|which)\b', utterance['text'].lower())),
+                len(utterance['text'].split()),
                 sum(1 for token in doc if token.pos_ == 'VERB'),
                 sum(1 for token in doc if token.pos_ == 'NOUN')
             ])
             X = scaler.transform([feat])
             role = 'Interviewer' if clf.predict(X)[0] == 0 else 'Interviewee'
-            results.append({**speaker, 'role': role})
+            results.append({**utterance, 'role': role})
         return results
     except Exception as e:
         logger.error(f"Role classification failed: {str(e)}")