app.py

import gradio as gr
from transformers import MarianMTModel, MarianTokenizer
from datetime import datetime
import langid
import os
import pyttsx3
import time
import warnings

# Optional: suppress sacremoses warning
warnings.filterwarnings("ignore", message="Recommended: pip install sacremoses.")

# Set FFmpeg path explicitly (for pydub and audio playback)
os.environ["PATH"] += os.pathsep + r"C:\ffmpeg\bin"

langid.set_languages(['en', 'fr', 'sw'])

MODEL_MAP = {
    "English → Swahili": "Helsinki-NLP/opus-mt-en-sw",
    "English → French": "Helsinki-NLP/opus-mt-en-fr",
    "French → English": "Helsinki-NLP/opus-mt-fr-en",
    "French → Swahili (via English)": ["Helsinki-NLP/opus-mt-fr-en", "Helsinki-NLP/opus-mt-en-sw"]
}

TONE_MODIFIERS = {
    "Neutral": "",
    "Romantic": "Express this romantically: ",
    "Formal": "Translate this in a formal tone: ",
    "Casual": "Make this sound casual: "
}

loaded_models = {}

def load_model(model_name):
    if model_name not in loaded_models:
        tokenizer = MarianTokenizer.from_pretrained(model_name)
        model = MarianMTModel.from_pretrained(model_name)
        loaded_models[model_name] = (tokenizer, model)
    return loaded_models[model_name]

def detect_language(text):
    try:
        lang, score = langid.classify(text)
        return lang
    except:
        return "unknown"

def translate(text, direction, tone):
    detected_lang = detect_language(text)
    expected_src = direction.split(" → ")[0].lower()

    if expected_src.startswith("english") and detected_lang != "en":
        warning = f"⚠ Detected language is '{detected_lang}', but you selected English as source."
    elif expected_src.startswith("french") and detected_lang != "fr":
        warning = f"⚠ Detected language is '{detected_lang}', but you selected French as source."
    elif expected_src.startswith("swahili") and detected_lang != "sw":
        warning = f"⚠ Detected language is '{detected_lang}', but you selected Swahili as source."
    else:
        warning = ""

    prompt = TONE_MODIFIERS[tone] + text
    model_info = MODEL_MAP[direction]

    if isinstance(model_info, list):
        tokenizer1, model1 = load_model(model_info[0])
        encoded1 = tokenizer1(prompt, return_tensors="pt", padding=True, truncation=True)
        intermediate = model1.generate(**encoded1)
        intermediate_text = tokenizer1.decode(intermediate[0], skip_special_tokens=True)

        tokenizer2, model2 = load_model(model_info[1])
        encoded2 = tokenizer2(intermediate_text, return_tensors="pt", padding=True, truncation=True)
        final = model2.generate(**encoded2)
        translation = tokenizer2.decode(final[0], skip_special_tokens=True)
    else:
        tokenizer, model = load_model(model_info)
        encoded = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True)
        generated = model.generate(**encoded)
        translation = tokenizer.decode(generated[0], skip_special_tokens=True)

    with open("translation_log.txt", "a", encoding="utf-8") as f:
        f.write(f"[{datetime.now()}] {direction} | Tone: {tone}\n")
        f.write(f"Input: {text}\nOutput: {translation}\n\n")

    return f"{warning}\n{translation}" if warning else translation

# Get available voices
engine = pyttsx3.init()
voices = engine.getProperty('voices')
voice_names = [voice.name for voice in voices]

def speak_text_to_file(text, voice_name):
    try:
        engine = pyttsx3.init()
        engine.setProperty('rate', 150)
        for voice in voices:
            if voice.name == voice_name:
                engine.setProperty('voice', voice.id)
                break
        output_path = "tts_output.wav"
        engine.save_to_file(text, output_path)
        engine.runAndWait()
        return output_path
    except Exception as e:
        return None

def transcribe_and_translate(audio_path, direction, tone):
    import speech_recognition as sr
    recognizer = sr.Recognizer()
    try:
        with sr.AudioFile(audio_path) as source:
            audio = recognizer.record(source)
        if len(audio.frame_data) < 10000:
            return "⚠ Audio too short or empty. Please try again."
        text = recognizer.recognize_google(audio)
        return translate(text, direction, tone)
    except Exception as e:
        return f"⚠ Could not transcribe audio: {e}"

# Gradio UI
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("## 🌍 EAC Translator")
    gr.Markdown("Supports English, French, and Swahili. Includes tone control, language detection, voice input, and speech playback.")

    with gr.Tabs():
        with gr.Tab("📝 Text Translation"):
            input_text = gr.Textbox(label="Text to Translate", lines=3)
            direction = gr.Dropdown(choices=list(MODEL_MAP.keys()), label="Translation Direction", value="English → Swahili")
            tone = gr.Radio(choices=list(TONE_MODIFIERS.keys()), label="Tone", value="Neutral")
            output_text = gr.Textbox(label="Translated Text", lines=3)
            voice_choice = gr.Dropdown(choices=voice_names, label="Voice for Playback", value=voice_names[0])
            audio_output = gr.Audio(label="Playback", interactive=False)
            translate_btn = gr.Button("Translate")
            speak_btn = gr.Button("🔊 Speak Translation")
            translate_btn.click(fn=translate, inputs=[input_text, direction, tone], outputs=output_text)
            speak_btn.click(fn=speak_text_to_file, inputs=[output_text, voice_choice], outputs=audio_output)

        with gr.Tab("🎙 Voice Translation"):
            audio_input = gr.Audio(sources=["microphone"], type="filepath", label="Speak Now")
            direction_voice = gr.Dropdown(choices=list(MODEL_MAP.keys()), label="Translation Direction", value="English → Swahili")
            tone_voice = gr.Radio(choices=list(TONE_MODIFIERS.keys()), label="Tone", value="Neutral")
            voice_output = gr.Textbox(label="Translated Text")
            voice_choice2 = gr.Dropdown(choices=voice_names, label="Voice for Playback", value=voice_names[0])
            audio_output2 = gr.Audio(label="Playback", interactive=False)
            voice_translate_btn = gr.Button("Transcribe & Translate")
            voice_speak_btn = gr.Button("🔊 Speak Translation")
            voice_translate_btn.click(fn=transcribe_and_translate, inputs=[audio_input, direction_voice, tone_voice], outputs=voice_output)
            voice_speak_btn.click(fn=speak_text_to_file, inputs=[voice_output, voice_choice2], outputs=audio_output2)

demo.launch()