app.py

import os
from gtts import gTTS
from io import BytesIO
from pydub import AudioSegment
from pydub.playback import play
import speech_recognition as sr
from IPython.display import Audio
import google.generativeai as genai
from dotenv import load_dotenv
import streamlit as st

# Path to store voice files
path = "../data/voice/"
os.makedirs(path, exist_ok=True)

# 1. Save and play voice created by Google Text-to-Speech (gTTS)
def text_to_audio(text, filename):
 tts = gTTS(text)
 file_path = os.path.join(path, filename)
 tts.save(file_path)
 return file_path

def play_audio(file_path):
 audio = AudioSegment.from_file(file_path)
 play(audio)

# 2. Use microphone to record voice
def record_audio(duration=4):
 recognizer = sr.Recognizer()
 with sr.Microphone() as source:
 print("Adjusting noise...")
 recognizer.adjust_for_ambient_noise(source, duration=1)
 print(f"Recording for {duration} seconds...")
 recorded_audio = recognizer.listen(source, timeout=duration)
 print("Done recording.")
 return recorded_audio

# 3. Convert the recorded voice to text through speech-to-text (STT)
def audio_to_text(audio):
 recognizer = sr.Recognizer()
 try:
 print("Recognizing the text...")
 text = recognizer.recognize_google(audio, language="en-US")
 print("Decoded Text: {}".format(text))
 except sr.UnknownValueError:
 text = "Google Speech Recognition could not understand the audio."
 except sr.RequestError:
 text = "Could not request results from Google Speech Recognition service."
 return text

# 4. Convert the text to voice through text-to-speech (TTS)
def text_to_speech(text):
 tts = gTTS(text)
 audio_buffer = BytesIO()
 tts.write_to_fp(audio_buffer)
 audio_buffer.seek(0)
 audio_segment = AudioSegment.from_file(audio_buffer, format="mp3")
 play(audio_segment)

# 5. Make a voice-to-voice stream
def voice_to_voice():
 recorded_audio = record_audio()
 text = audio_to_text(recorded_audio)
 text_to_speech(text)

# 6. Integrate an LLM to respond to voice input with voice output
load_dotenv()
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
genai.configure(api_key=GOOGLE_API_KEY)
gemini_pro = genai.GenerativeModel(model_name="models/gemini-pro")

def respond_by_gemini(input_text, role_text, instructions_text):
 final_prompt = [
 "ROLE: " + role_text,
 "INPUT_TEXT: " + input_text,
 instructions_text,
 ]
 response = gemini_pro.generate_content(
 final_prompt,
 stream=True,
 )
 response_list = []
 for chunk in response:
 response_list.append(chunk.text)
 response_text = "".join(response_list)
 return response_text

def llm_voice_response():
 role = 'You are an intelligent assistant to chat on the topic: `{}`.'
 topic = 'The future of artificial intelligence'
 role_text = role.format(topic)
 instructions = 'Respond to the INPUT_TEXT briefly in chat style. Respond based on your knowledge about `{}` in brief chat style.'
 instructions_text = instructions.format(topic)
 
 recorded_audio = record_audio()
 text = audio_to_text(recorded_audio)
 response_text = text
 if text not in ["Google Speech Recognition could not understand the audio.", "Could not request results from Google Speech Recognition service."]:
 response_text = respond_by_gemini(text, role_text, instructions_text)
 text_to_speech(response_text)

# 7. Build a Web interface for the LLM-supported voice assistant
def main():
 # Streamlit setup with custom CSS
 st.set_page_config(page_title="LLM-Supported Voice Assistant", layout="wide")
 
 st.markdown("""
 <link href="https://stackpath.bootstrapcdn.com/bootstrap/4.5.2/css/bootstrap.min.css" rel="stylesheet">
 <style>
 .main {background-color: #f5f5f5;}
 .container {max-width: 800px; margin: auto; padding-top: 50px;}
 .title {font-family: 'Arial', sans-serif; color: #333333; margin-bottom: 30px;}
 .btn {background-color: #4CAF50; color: white; border: none; padding: 10px 20px; cursor: pointer; font-size: 16px;}
 .btn:hover {background-color: #45a049;}
 </style>
 """, unsafe_allow_html=True)
 
 st.markdown("<div class='container'><h1 class='title'>LLM-Supported Voice Assistant</h1></div>", unsafe_allow_html=True)
 
 st.write("This is a voice assistant with LLM support. Speak to the microphone, and the assistant will respond.")
 
 if st.button("Record and Get Response", key="record_btn"):
 st.write("Listening...")
 llm_voice_response()
 st.write("Done.")
 
 st.markdown("<div class='container'><h5>Press the button and speak to the microphone. The assistant will generate a response based on the input and speak it out loud.</h5></div>", unsafe_allow_html=True)

if __name__ == "__main__":
 main()