app.py

import streamlit as st
from transformers import MarianMTModel, MarianTokenizer
import whisper
import tempfile
import base64
import os
from audiorecorder import audiorecorder  # pip install streamlit-audiorecorder

# Load models once
@st.cache_resource
def load_models():
    whisper_model = whisper.load_model("base")
    tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
    translator = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
    return whisper_model, tokenizer, translator

st.title("🎙️ Live Arabic Sermon Translator")

st.markdown("Click the mic, say something in Arabic, and wait a few seconds for translation.")

# Record audio
audio = audiorecorder("Start Recording", "Stop Recording")

if len(audio) > 0:
    st.audio(audio.tobytes(), format="audio/wav")

    # Save audio to temp file
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
        f.write(audio.tobytes())
        temp_wav_path = f.name

    st.info("Transcribing Arabic...")
    whisper_model, tokenizer, translator = load_models()
    transcription = whisper_model.transcribe(temp_wav_path, language="ar")
    arabic_text = transcription["text"]
    st.markdown("### Arabic")
    st.write(arabic_text)

    st.info("Translating to English...")
    tokens = tokenizer(arabic_text, return_tensors="pt", padding=True)
    output = translator.generate(**tokens)
    english_text = tokenizer.decode(output[0], skip_special_tokens=True)

    st.markdown("### English")
    st.success(english_text)

    os.remove(temp_wav_path)