Update app.py

app.py

@@ -1,180 +1,159 @@
-import os
-os.system("pip install --upgrade openai-whisper torch")
-os.system("pip install --upgrade transformers")
 import whisper
 import gradio as gr
 import torch
-from transformers import BertTokenizer, BertForSequenceClassification
+from transformers import BertTokenizer, BertForSequenceClassification, pipeline
 from app.questions import get_question
 
-# Load Whisper model
+# Load models
 whisper_model = whisper.load_model("small")
-
-# Load the pre-trained model and tokenizer for confidence analysis
-confidence_model = BertForSequenceClassification.from_pretrained('RiteshAkhade/Confidence')
-confidence_tokenizer = BertTokenizer.from_pretrained('RiteshAkhade/Confidence')
-
-# Load pre-trained context analysis model (BERT-based)
+confidence_model = BertForSequenceClassification.from_pretrained('/home/ghost/LLM/confidence_model1')
+confidence_tokenizer = BertTokenizer.from_pretrained('/home/ghost/LLM/confidence_tokenizer1')
 context_model = BertForSequenceClassification.from_pretrained('RiteshAkhade/context_model')
 context_tokenizer = BertTokenizer.from_pretrained('RiteshAkhade/context_model')
-
-# Function to perform context analysis using the BERT model
+emotion_pipe = pipeline("text-classification", model="bhadresh-savani/distilbert-base-uncased-emotion", top_k=1)
+
+# Emotion map with labels and emojis
+interview_emotion_map = {
+    "joy": ("Confident", "🙂"),
+    "fear": ("Nervous", "😨"),
+    "sadness": ("Uncertain", "🙁"),
+    "anger": ("Frustrated", "😠"),
+    "surprise": ("Curious", "😮"),
+    "neutral": ("Calm", "😐"),
+    "disgust": ("Disengaged", "😒"),
+}
+
+# Static question sets
+technical_questions = [get_question(i) for i in range(6)]
+non_technical_questions = [
+    "Tell me about yourself.",
+    "What are your strengths and weaknesses?",
+    "Where do you see yourself in 5 years?",
+    "How do you handle stress or pressure?",
+    "Describe a time you faced a conflict and how you resolved it.",
+    "What motivates you to do your best?"
+]
+
+# Index trackers
+current_tech_index = 0
+current_non_tech_index = 0
+
+# Relevance prediction
 def predict_relevance(question, answer):
-    if not answer.strip():  # Check for empty answers
+    if not answer.strip():
         return "Irrelevant"
-    
     inputs = context_tokenizer(question, answer, return_tensors="pt", padding=True, truncation=True)
     context_model.eval()
-
     with torch.no_grad():
         outputs = context_model(**inputs)
-        logits = outputs.logits
-        probabilities = torch.softmax(logits, dim=-1)
-        threshold = 0.5
-        relevant_prob = probabilities[0, 1]  # Probability for relevant class
-    
-    if relevant_prob > threshold:
-        return "Relevant"
-    else:
-        return "Irrelevant"
+        probabilities = torch.softmax(outputs.logits, dim=-1)
+        return "Relevant" if probabilities[0, 1] > 0.5 else "Irrelevant"
 
-#  confidence analysis
+# Confidence prediction
 def predict_confidence(question, answer, threshold=0.4):
-    if not isinstance(answer, str) or not answer.strip():  
+    if not isinstance(answer, str) or not answer.strip():
         return "Not Confident"
-
-    # Tokenize input
     inputs = confidence_tokenizer(question, answer, return_tensors="pt", padding=True, truncation=True)
-    
-    # Set model to evaluation mode
     confidence_model.eval()
-
     with torch.no_grad():
         outputs = confidence_model(**inputs)
-        logits = outputs.logits
-        probabilities = torch.softmax(logits, dim=-1)
-
-        confidence_prob = probabilities[0, 1].item()  # Extract probability for "Confident" class
-
-        
-        return "Confident" if confidence_prob > threshold else "Not Confident"
-
-# Questions from questions.py
-def fetch_questions():
-    return [get_question(i) for i in range(6)]
-
-questions = fetch_questions()
-current_question_index = 0
-
-# Function to show the current question
-def show_question():
-    global current_question_index
-    question = questions[current_question_index]
-    return question
-
-# Function to move to the next question
-def next_question():
-    global current_question_index
-    current_question_index = (current_question_index + 1) % len(questions)
-    return show_question(), None, "", "", ""
-
-# Function to transcribe the audio input and perform both context and confidence analyses
-def transcribe_and_analyze(audio, question):
+        probabilities = torch.softmax(outputs.logits, dim=-1)
+        return "Confident" if probabilities[0, 1].item() > threshold else "Not Confident"
+
+# Emotion detection
+def detect_emotion(answer):
+    if not answer.strip():
+        return "No Answer", ""
+    result = emotion_pipe(answer)
+    label = result[0][0]["label"].lower()
+    emotion_text, emoji = interview_emotion_map.get(label, ("Unknown", "❓"))
+    return emotion_text, emoji
+
+# Question navigation (non-tech)
+def show_non_tech_question():
+    global current_non_tech_index
+    return non_technical_questions[current_non_tech_index]
+
+def next_non_tech_question():
+    global current_non_tech_index
+    current_non_tech_index = (current_non_tech_index + 1) % len(non_technical_questions)
+    return non_technical_questions[current_non_tech_index], None, "", ""
+
+# Question navigation (tech)
+def show_tech_question():
+    global current_tech_index
+    return technical_questions[current_tech_index]
+
+def next_tech_question():
+    global current_tech_index
+    current_tech_index = (current_tech_index + 1) % len(technical_questions)
+    return technical_questions[current_tech_index], None, "", "", ""
+
+# Transcribe + analyze (non-technical)
+def transcribe_and_analyze_non_tech(audio, question):
     try:
-        # Load and process audio using Whisper
         audio = whisper.load_audio(audio)
         audio = whisper.pad_or_trim(audio)
         mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)
-        options = whisper.DecodingOptions(fp16=False)
-        result = whisper.decode(whisper_model, mel, options)
-        
-        # Get the transcribed text
+        result = whisper.decode(whisper_model, mel, whisper.DecodingOptions(fp16=False))
         transcribed_text = result.text
+        emotion_text, emoji = detect_emotion(transcribed_text)
+        return transcribed_text, f"{emotion_text} {emoji}"
+    except Exception as e:
+        return f"Error: {str(e)}", "❓"
 
-        # Perform context and confidence analysis
+# Transcribe + analyze (technical)
+def transcribe_and_analyze_tech(audio, question):
+    try:
+        audio = whisper.load_audio(audio)
+        audio = whisper.pad_or_trim(audio)
+        mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)
+        result = whisper.decode(whisper_model, mel, whisper.DecodingOptions(fp16=False))
+        transcribed_text = result.text
         context_result = predict_relevance(question, transcribed_text)
         confidence_result = predict_confidence(question, transcribed_text)
-
-        # Return the results
         return transcribed_text, context_result, confidence_result
     except Exception as e:
         return f"Error: {str(e)}", "", ""
 
-#  Gradio interface 
-with gr.Blocks() as demo:
-    
-    gr.HTML('''
-    <style>
-        body { 
-            background-color: #f0f0f0; 
-        }
-        #title { 
-            color: grey; 
-            font-size: 30px; 
-            text-align: center; 
-            margin-bottom: 20px; 
-        }
-        .transcribe-btn, .next-btn { 
-            background-color: #4CAF50; 
-            color: white; 
-            font-size: 16px; 
-            padding: 10px 20px; 
-            border-radius: 5px; 
-            cursor: pointer; 
-            margin-top: 10px;
-        }
-        .transcribe-btn:hover, .next-btn:hover { 
-            background-color: #45a049; 
-        }
-        #question-box { 
-            font-size: 20px; 
-            color: #555; 
-            text-align: center; 
-        }
-        #text-box {
-            font-size: 18px;
-            color: #333;
-        }
-        #context-box, #confidence-box {
-            font-size: 18px;
-            color: #333;
-        }
-    </style>
-    ''')
-
-    # Title
-    gr.Markdown("<h1 id='title'>INTERVIEW PREPARATION MODEL</h1>")
-
-    # Question display
-    with gr.Row():
-        question_display = gr.Textbox(label="Interview Question", value=show_question(), interactive=False, elem_id="question-box")
-
-    # Audio input and transcription section
-    with gr.Row():
-        audio_input = gr.Audio(type="filepath", label="Record Your Answer")
-
-    # Separate text boxes for the transcribed text, context, and confidence analysis
-    with gr.Row():
-        transcribed_text = gr.Textbox(label="Your Answer (Transcription)", interactive=False, lines=5, elem_id="text-box")
-    
-    with gr.Row():
-        context_analysis_result = gr.Textbox(label="Context Analysis", interactive=False, elem_id="context-box")
-    
-    with gr.Row():
-        confidence_analysis_result = gr.Textbox(label="Confidence Analysis", interactive=False, elem_id="confidence-box")
-    
-    # Automatically transcribe the audio and analyze context and confidence when audio is provided
-    audio_input.change(fn=transcribe_and_analyze, 
-                       inputs=[audio_input, question_display], 
-                       outputs=[transcribed_text, context_analysis_result, confidence_analysis_result])
-
-    # Button to get the next question
-    with gr.Row():    
-        next_button = gr.Button("Next Question", elem_classes="next-btn")
+# UI layout
+with gr.Blocks(css="textarea, .gr-box { font-size: 18px !important; }") as demo:
+    gr.HTML("<h1 style='text-align: center; font-size: 32px;'>INTERVIEW PREPARATION MODEL</h1>")
+
+    with gr.Tabs():
         
-        # Clear audio, transcription, context, and confidence when moving to the next question
-        next_button.click(next_question, 
-                          outputs=[question_display, audio_input, transcribed_text, context_analysis_result, confidence_analysis_result])
+        # NON-TECHNICAL TAB
+        with gr.Tab("Non-Technical"):
+            gr.Markdown("### Emotional Context Analysis (🧠 + 😊)")
+            question_display_1 = gr.Textbox(label="Interview Question", value=show_non_tech_question(), interactive=False)
+            audio_input_1 = gr.Audio(type="filepath", label="Record Your Answer")
+            transcribed_text_1 = gr.Textbox(label="Transcribed Answer", interactive=False, lines=4)
+            emotion_output = gr.Textbox(label="Detected Emotion", interactive=False)
+
+            audio_input_1.change(fn=transcribe_and_analyze_non_tech,
+                                 inputs=[audio_input_1, question_display_1],
+                                 outputs=[transcribed_text_1, emotion_output])
+
+            next_button_1 = gr.Button("Next Question")
+            next_button_1.click(fn=next_non_tech_question,
+                                outputs=[question_display_1, audio_input_1, transcribed_text_1, emotion_output])
+
+        # TECHNICAL TAB
+        with gr.Tab("Technical"):
+            gr.Markdown("### Technical Question Analysis (🎓 + 🤖)")
+            question_display_2 = gr.Textbox(label="Interview Question", value=show_tech_question(), interactive=False)
+            audio_input_2 = gr.Audio(type="filepath", label="Record Your Answer")
+            transcribed_text_2 = gr.Textbox(label="Transcribed Answer", interactive=False, lines=4)
+            context_analysis_result = gr.Textbox(label="Context Analysis", interactive=False)
+            confidence_analysis_result = gr.Textbox(label="Confidence Analysis", interactive=False)
+
+            audio_input_2.change(fn=transcribe_and_analyze_tech,
+                                 inputs=[audio_input_2, question_display_2],
+                                 outputs=[transcribed_text_2, context_analysis_result, confidence_analysis_result])
+
+            next_button_2 = gr.Button("Next Question")
+            next_button_2.click(fn=next_tech_question,
+                                outputs=[question_display_2, audio_input_2, transcribed_text_2,
+                                         context_analysis_result, confidence_analysis_result])
 
-# Launch app
 demo.launch(share=True)