chatbot_ui1.py

import streamlit as st
import pandas as pd
import torch
from io import BytesIO
from gtts import gTTS
from sentence_transformers import SentenceTransformer, util
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import datetime  # Logging
import json  # Chat history
from textblob import TextBlob  # Sentiment analysis
from langdetect import detect  # Language detection
from googletrans import Translator  # FIXED: Import Translator
import speech_recognition as sr  # Voice recognition

# Load dataset
@st.cache_data
def load_dataset():
    df = pd.read_csv("Mental_Health_FAQ.csv")
    if "Questions" not in df.columns or "Answers" not in df.columns:
        st.error("❌ Error: Dataset must contain 'Questions' and 'Answers' columns.")
        return [], []
    
    return df["Questions"].dropna().tolist(), df["Answers"].dropna().tolist()

questions, answers = load_dataset()

# Load models
@st.cache_resource
def load_models():
    embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
    chatbot_model_name = "facebook/blenderbot-400M-distill"
    chatbot_model = AutoModelForSeq2SeqLM.from_pretrained(chatbot_model_name)
    chatbot_tokenizer = AutoTokenizer.from_pretrained(chatbot_model_name)
    return embedding_model, chatbot_model, chatbot_tokenizer

embedding_model, chatbot_model, chatbot_tokenizer = load_models()

# Generate embeddings for dataset questions
@st.cache_data
def generate_question_embeddings():
    if questions:
        return embedding_model.encode(questions, convert_to_tensor=True)
    return None

question_embeddings = generate_question_embeddings()

def get_best_response(user_input):
    """Finds the closest matching FAQ answer or generates a response using BlenderBot."""
    if not questions:
        return "Sorry, the FAQ database is empty."

    input_embedding = embedding_model.encode(user_input, convert_to_tensor=True)
    similarities = util.pytorch_cos_sim(input_embedding, question_embeddings)[0].cpu()

    best_match_idx = torch.argmax(similarities).item()
    best_match_score = similarities[best_match_idx].item()

    if best_match_score > 0.7:
        return answers[best_match_idx]
    
    # Generate response using BlenderBot
    inputs = chatbot_tokenizer(user_input, return_tensors="pt")
    outputs = chatbot_model.generate(**inputs)
    return chatbot_tokenizer.decode(outputs[0], skip_special_tokens=True)

def analyze_sentiment(text):
    """Analyzes sentiment and returns an emoji representation."""
    sentiment = TextBlob(text).sentiment.polarity
    if sentiment > 0:
        return "😊 Positive"
    elif sentiment < 0:
        return "😞 Negative"
    else:
        return "😐 Neutral"

def text_to_speech(text):
    """Converts chatbot response to speech."""
    tts = gTTS(text=text, lang="en")
    audio_file = BytesIO()
    tts.write_to_fp(audio_file)
    audio_file.seek(0)
    return audio_file

def transcribe_audio():
    """Allows user to speak input instead of typing."""
    recognizer = sr.Recognizer()
    with sr.Microphone() as source:
        st.write("🎤 Listening... Speak now!")
        audio = recognizer.listen(source)
    
    try:
        return recognizer.recognize_google(audio)
    except sr.UnknownValueError:
        return "Sorry, I couldn't understand that."
    except sr.RequestError:
        return "Speech recognition service error."

# Initialize Translator (FIXED)
translator = Translator()

# Streamlit UI
st.title("🧠 Mental Health Chatbot 🤖")
st.write("Ask me anything about mental health!")

# User Input
st.subheader("💬 Chat with me")
user_input = st.chat_input("Type your message here...")

# Voice Input Button
if st.button("🎙️ Use Voice Input"):
    user_input = transcribe_audio()

# Process User Input
if user_input:
    # Detect language using `langdetect`
    detected_lang = detect(user_input)
    
    if detected_lang != "en":
        user_input = translator.translate(user_input, dest="en").text  # FIXED

    # Get chatbot response
    response = get_best_response(user_input)

    # Sentiment analysis
    sentiment_result = analyze_sentiment(user_input)

    # Convert response to speech
    audio_file = text_to_speech(response)

    # Display chatbot response
    st.write(f"**Response:** {response}")
    st.write(f"**Sentiment:** {sentiment_result}")
    st.audio(audio_file, format="audio/mp3", start_time=0)