app.py

import streamlit as st
import whisper
from transformers import pipeline
import speech_recognition as sr
from gtts import gTTS
import tempfile
import os
import base64

# Load Whisper model for Speech-to-Text
@st.cache_resource
def load_whisper_model():
    return whisper.load_model("base")  # You can use other models like "large" for better accuracy

# Load translation model from Hugging Face
@st.cache_resource
def load_translation_model(input_lang, output_lang):
    model_name = f"Helsinki-NLP/opus-mt-{input_lang}-{output_lang}"
    return pipeline("translation", model=model_name)

# Initialize speech recognition
recognizer = sr.Recognizer()

def speech_to_text(audio_file):
    with sr.AudioFile(audio_file) as source:
        audio_data = recognizer.record(source)
        return recognizer.recognize_google(audio_data)

def translate_text(input_text, translation_pipeline):
    return translation_pipeline(input_text)[0]["translation_text"]

def text_to_speech(text, language):
    tts = gTTS(text=text, lang=language)
    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
    tts.save(temp_file.name)
    return temp_file.name

# Streamlit UI
st.title("Voice-to-Voice Translator")

# Sidebar for user input
st.sidebar.header("Settings")
input_lang = st.sidebar.selectbox("Select Input Language", ["en", "fr", "es", "de", "hi"])
output_lang = st.sidebar.selectbox("Select Output Language", ["en", "fr", "es", "de", "hi"])

# Initialize models
whisper_model = load_whisper_model()
translation_pipeline = load_translation_model(input_lang, output_lang)

# Input options
st.header("Input your voice or text")

# Option to upload an audio file
audio_file = st.file_uploader("Upload an audio file (WAV format)", type=["wav"])

if audio_file is not None:
    # Convert speech to text
    input_text = speech_to_text(audio_file)
    st.write(f"Recognized Text: {input_text}")
    
    # Translate the text
    translated_text = translate_text(input_text, translation_pipeline)
    st.success(f"Translated Text: {translated_text}")

    # Convert translated text to speech
    output_audio = text_to_speech(translated_text, output_lang)
    audio_bytes = open(output_audio, "rb").read()
    
    # Play the audio
    st.audio(audio_bytes, format="audio/mp3")
    
    # Provide download link for audio
    b64 = base64.b64encode(audio_bytes).decode()
    href = f'<a href="data:audio/mp3;base64,{b64}" download="translated_audio.mp3">Download Translated Audio</a>'
    st.markdown(href, unsafe_allow_html=True)

else:
    st.write("Please upload an audio file to get started.")