app.py

import streamlit as st
from audio_recorder_streamlit import audio_recorder
from groq import Groq
import os
from langchain_groq import ChatGroq
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
import edge_tts
import asyncio
from dotenv import load_dotenv
load_dotenv()

# Page config
st.set_page_config(page_title="Voice AI Assistant", page_icon="🎤", layout="centered")

# Theme toggle
if "dark_mode" not in st.session_state:
    st.session_state.dark_mode = False  # default: light mode
dm = st.sidebar.checkbox("🌙 Dark Mode", value=st.session_state.dark_mode)
st.session_state.dark_mode = dm

# Theme colors
BG = "#0f1620" if dm else "#f8f9fa"
PANEL = "#1c2330" if dm else "#ffffff"
TEXT = "#e3e8f1" if dm else "#1a1a1a"
CARD = "#2a3240" if dm else "#f1f3f5"
ACCENT = "#ff5252"
BORDER = "#333" if dm else "#ddd"

# Custom CSS
st.markdown(f"""
<style>
    .stApp {{
        background-color: {BG};
        color: {TEXT};
    }}
    [data-testid="stSidebar"] {{
        background-color: {PANEL};
    }}
    .block-container {{
        padding-top: 2rem;
        padding-bottom: 2rem;
    }}
    h1, h2, h3, h4 {{
        color: {TEXT};
    }}
    .conversation-block {{
        background-color: {CARD};
        padding: 1rem;
        border-radius: 8px;
        margin-bottom: 1rem;
        border: 1px solid {BORDER};
    }}
    .question {{
        font-weight: bold;
        color: {ACCENT};
    }}
    .answer {{
        margin-top: 0.5rem;
        color: {TEXT};
    }}
    .audio-player {{
        margin-top: 0.5rem;
    }}
    .status-bar {{
        font-style: italic;
        color: {TEXT}AA;
        margin-bottom: 1rem;
    }}
</style>
""", unsafe_allow_html=True)

# App UI
st.title("🎤 Voice AI Assistant")

# Session init
if "conversation" not in st.session_state:
    st.session_state.conversation = []  # list of (question, answer, audio_filename)
if "audio_count" not in st.session_state:
    st.session_state.audio_count = 1

status = st.empty()
status.markdown("<div class='status-bar'>🎙️ Press mic button or type to ask a question</div>", unsafe_allow_html=True)

recorded_audio = audio_recorder(sample_rate=8000)
text_input = st.chat_input("Type your question here...")

# ----- INPUT HANDLER -----
def handle_input(user_text):
    status.markdown("<div class='status-bar'>🤖 Thinking...</div>", unsafe_allow_html=True)
    response = answer(user_text)
    audio_file = f"output{st.session_state.audio_count}.wav"
    status.markdown("<div class='status-bar'>🎧 Converting response to audio...</div>", unsafe_allow_html=True)
    asyncio.run(convert_audio(response, audio_file))
    st.session_state.audio_count += 1

    st.session_state.conversation.append((f"Q: {user_text}", f"A: {response}", audio_file))
    status.markdown("<div class='status-bar'>✅ Ask another question...</div>", unsafe_allow_html=True)

# ----- PROCESS INPUT -----
if text_input:
    handle_input(text_input)
elif recorded_audio:
    status.markdown("<div class='status-bar'>🧠 Transcribing speech...</div>", unsafe_allow_html=True)
    data_to_file(recorded_audio)
    transcription = audio_to_text("temp_audio.wav")
    handle_input(transcription)

# ----- SHOW CONVERSATION -----
if st.session_state.conversation:
    st.markdown("## 🧾 Conversation History")
    for i, (q, a, audio_path) in enumerate(st.session_state.conversation):
        with st.container():
            st.markdown(f"<div class='conversation-block'>", unsafe_allow_html=True)
            st.markdown(f"<div class='question'>{q}</div>", unsafe_allow_html=True)
            st.markdown(f"<div class='answer'>{a}</div>", unsafe_allow_html=True)
            st.audio(audio_path, format="audio/wav", autoplay=(i == len(st.session_state.conversation)-1))
            st.markdown("</div>", unsafe_allow_html=True)

# ----- AUDIO TO TEXT -----
def data_to_file(audio_blob):
    with open("temp_audio.wav", "wb") as f:
        f.write(audio_blob)

def audio_to_text(path):
    client = Groq(api_key=os.getenv("GROQ_API_KEY"))
    with open(path, "rb") as f:
        transcription = client.audio.translations.create(
            file=(path, f.read()),
            model='whisper-large-v3',
        )
    return transcription.text

# ----- LLM ANSWER -----
def answer(question):
    model = ChatGroq(model="llama-3.3-70b-versatile", temperature=0.6)
    prompt = ChatPromptTemplate([
        ("system", "You are a knowledgeable AI assistant. Keep answers clear, brief, and well-punctuated for speech conversion."),
        ("user", "User Query: {question}")
    ])
    parser = StrOutputParser()
    chain = prompt | model | parser
    return chain.invoke({'question': question})

# ----- TEXT TO AUDIO -----
async def convert_audio(text, filename):
    voice = "fr-FR-VivienneMultilingualNeural"
    communicate = edge_tts.Communicate(text, voice)
    await communicate.save(filename)