Update app.py

app.py

@@ -3,6 +3,7 @@ from transformers import pipeline
 from speechbrain.pretrained import Tacotron2, HIFIGAN, EncoderDecoderASR
 import matplotlib.pyplot as plt
 import pandas as pd
+import random
 
 # Initialize psychometric model
 psych_model_name = "KevSun/Personality_LM"
@@ -13,18 +14,6 @@ asr_model = EncoderDecoderASR.from_hparams(source="speechbrain/asr-crdnn-rnnlm-l
 tts_model = Tacotron2.from_hparams(source="speechbrain/tts-tacotron2-ljspeech", savedir="tmp_tts")
 voc_model = HIFIGAN.from_hparams(source="speechbrain/tts-hifigan-ljspeech", savedir="tmp_voc")
 
-# Psychometric Test Questions
-text_questions = [
-    "How do you handle criticism?",
-    "Describe a time when you overcame a challenge.",
-    "What motivates you to work hard?"
-]
-
-audio_questions = [
-    "What does teamwork mean to you?",
-    "How do you handle stressful situations?"
-]
-
 # Function to analyze text responses
 def analyze_text_responses(responses):
     analysis = [psych_model(response)[0] for response in responses]
@@ -37,6 +26,7 @@ def generate_audio_question(question):
     waveforms = voc_model.decode_batch(mel_output)
     return waveforms[0].numpy()
 
+# Function to process audio response
 def process_audio_response(audio):
     # Check if the audio input is None
     if audio is None:
@@ -49,63 +39,49 @@ def process_audio_response(audio):
     except Exception as e:
         return f"Error processing audio: {str(e)}"
 
-# Gradio interface functions
-def text_part(candidate_name, responses):
-    traits = analyze_text_responses(responses)
-    df = pd.DataFrame(traits.items(), columns=["Trait", "Score"])
-    plt.figure(figsize=(8, 6))
-    plt.bar(df["Trait"], df["Score"], color="skyblue")
-    plt.title(f"Psychometric Analysis for {candidate_name}")
-    plt.xlabel("Traits")
-    plt.ylabel("Score")
-    plt.xticks(rotation=45)
-    plt.tight_layout()
-    return df, plt
-
-def audio_part(candidate_name, audio_responses):
-    # Check if any audio response is invalid (None)
-    valid_audio_responses = [process_audio_response(audio) for audio in audio_responses if audio is not None]
-    
-    # If all responses are invalid, return an error message
-    if not valid_audio_responses:
-        return "No valid audio responses provided", None
+# Function to generate dynamic questions based on answers
+def generate_dynamic_question(previous_answer):
+    # Example of simple follow-up questions based on the answer
+    if "teamwork" in previous_answer.lower():
+        return "Can you share a specific instance where you worked in a team?"
+    elif "challenge" in previous_answer.lower():
+        return "How did you overcome that challenge? What steps did you take?"
+    elif "stress" in previous_answer.lower():
+        return "How do you manage stress during high-pressure situations?"
+    else:
+        # Default follow-up question
+        return "Can you tell me more about that?"
 
-    traits = analyze_text_responses(valid_audio_responses)
-    df = pd.DataFrame(traits.items(), columns=["Trait", "Score"])
-    plt.figure(figsize=(8, 6))
-    plt.bar(df["Trait"], df["Score"], color="lightcoral")
-    plt.title(f"Audio Psychometric Analysis for {candidate_name}")
-    plt.xlabel("Traits")
-    plt.ylabel("Score")
-    plt.xticks(rotation=45)
-    plt.tight_layout()
-    return df, plt
-
-
-# Gradio UI function
+# Gradio UI function to handle dynamic conversation
 def chat_interface(candidate_name, *responses):
-    # Separate text responses and audio responses
-    num_text_questions = len(text_questions)
-    text_responses = responses[:num_text_questions]
-    audio_responses = responses[num_text_questions:]
+    conversation_history = []
+    
+    # Iterate through responses to generate follow-up questions
+    for i, response in enumerate(responses):
+        conversation_history.append(f"Q{i+1}: {response}")
+        
+        # Generate dynamic question based on the previous response
+        dynamic_question = generate_dynamic_question(response)
+        conversation_history.append(f"Follow-up Question: {dynamic_question}")
     
     # Process text responses
-    text_df, text_plot = text_part(candidate_name, text_responses)
+    text_df, text_plot = text_part(candidate_name, responses)
     
     # Process audio responses
-    audio_df, audio_plot = audio_part(candidate_name, audio_responses)
+    audio_df, audio_plot = audio_part(candidate_name, responses)
     
-    return text_df, text_plot, audio_df, audio_plot
+    # Return conversation history and analysis
+    return "\n".join(conversation_history), text_df, text_plot, audio_df, audio_plot
 
 # Create text inputs and audio inputs
-text_inputs = [gr.Textbox(label=f"Response to Q{i+1}: {q}") for i, q in enumerate(text_questions)]
-audio_inputs = [gr.Audio(label=f"Response to Q{i+1}: {q}", type="filepath") for i, q in enumerate(audio_questions)]
+text_inputs = [gr.Textbox(label=f"Response to Q{i+1}:") for i in range(5)]  # Assuming we have up to 5 text responses
+audio_inputs = [gr.Audio(label=f"Response to Audio Q{i+1}:") for i in range(2)]  # Assuming we have up to 2 audio responses
 
 interface = gr.Interface(
     fn=chat_interface,
     inputs=[gr.Textbox(label="Candidate Name")] + text_inputs + audio_inputs,
-    outputs=["dataframe", "plot", "dataframe", "plot"],
-    title="Psychometric Analysis Chatbot"
+    outputs=["text", "dataframe", "plot", "dataframe", "plot"],
+    title="Dynamic Psychometric Analysis Chatbot"
 )
 
 # Launch the interface