app.py

import streamlit as st
from audio_recorder_streamlit import audio_recorder
import speech_recognition as sr
from gtts import gTTS
import tempfile
import os
from anthropic import Client  # For Claude Haiku model

# Claude API setup
CLAUDE_API_KEY = st.secrets['ANTHROPIC_API_KEY']  # Store your Claude API key in Streamlit secrets
client = Client(api_key=CLAUDE_API_KEY)

# Main function for chatbot app
def main():
    st.title("🎤 اردو وائس چیٹ بوٹ")
    
    # Sidebar with information
    st.sidebar.title("حامش راج")
    st.sidebar.write("ماہر ڈیٹا سائنس اور جنریٹو اے آئی")
    
    st.markdown("**اپنی آواز ریکارڈ کریں اور جواب حاصل کریں**")
    
    # Audio Recorder
    audio_data = audio_recorder(text='اردو میں بولیئے', icon_size="2x", icon_name="microphone-lines", key="urdu_recorder")
    
    if audio_data is not None:
        # Save the recorded audio to a temporary file
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_audio_file:
            temp_audio_file.write(audio_data)
            temp_audio_file_path = temp_audio_file.name

        # Convert audio to text (Speech to Text in Urdu)
        user_input_text = convert_audio_to_text(temp_audio_file_path)

        # Display user input text
        st.write(f"**آپ نے کہا:** {user_input_text}")
        
        # Get LLM (Claude) response
        response_text = get_claude_response(user_input_text)

        # Display chatbot's text response
        st.write(f"**جواب:** {response_text}")

        # Convert response text to audio and play it
        response_audio = convert_text_to_audio(response_text)
        st.audio(response_audio)

        # Clean up temporary audio file
        os.remove(temp_audio_file_path)

# Function to convert audio to text (Urdu Speech Recognition)
def convert_audio_to_text(audio_file_path):
    recognizer = sr.Recognizer()
    with sr.AudioFile(audio_file_path) as source:
        audio_data = recognizer.record(source)
        try:
            text = recognizer.recognize_google(audio_data, language="ur")
            return text
        except sr.UnknownValueError:
            return "معذرت، میں آپ کی آواز سمجھ نہیں سکا"
        except sr.RequestError:
            return "معذرت، سرور دستیاب نہیں ہے"

# Function to get response from Claude (Langchain with RAG)
def get_claude_response(prompt_text):
    prompt = f"براہ کرم اردو میں جواب دیں: {prompt_text}"
    try:
        # Query Claude Haiku via Langchain
        response = client.completions.create(
            model="claude-v1",  # Claude Haiku model
            prompt=prompt,
            max_tokens_to_sample=100,
        )
        return response['completion']
    except Exception as e:
        return f"خطا: {str(e)}"

# Function to convert text to Urdu audio (Text-to-Speech)
def convert_text_to_audio(text):
    try:
        tts = gTTS(text=text, lang='ur')
        temp_audio_path = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False).name
        tts.save(temp_audio_path)
        return temp_audio_path
    except Exception as e:
        return None

if __name__ == "__main__":
    main()