app.py

import gradio as gr
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"
psych_model = pipeline("text-classification", model=psych_model_name)

# Initialize ASR and TTS models
asr_model = EncoderDecoderASR.from_hparams(source="speechbrain/asr-crdnn-rnnlm-librispeech", savedir="tmp_asr")
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")

# Function to analyze text responses
def analyze_text_responses(responses):
    analysis = [psych_model(response)[0] for response in responses]
    traits = {response["label"]: response["score"] for response in analysis}
    return traits

# Function to handle TTS
def generate_audio_question(question):
    mel_output, alignment, _ = tts_model.encode_text(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:
        return "No audio provided"
    
    # Process the audio if it's a valid input
    try:
        text_response = asr_model.transcribe_file(audio)
        return text_response
    except Exception as e:
        return f"Error processing audio: {str(e)}"

# 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?"

# Gradio UI function to handle dynamic conversation
def chat_interface(candidate_name, *responses):
    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, responses)
    
    # Process audio responses
    audio_df, audio_plot = audio_part(candidate_name, responses)
    
    # 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}:") 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=["text", "dataframe", "plot", "dataframe", "plot"],
    title="Dynamic Psychometric Analysis Chatbot"
)

# Launch the interface
interface.launch()