import gradio as gr
from gtts import gTTS
import os
import uuid
import tempfile
from transformers import pipeline
from deep_translator import GoogleTranslator
import fasttext

# Load language detection model
lang_model = fasttext.load_model("lid.176.bin")

# Load Hugging Face model
qa_pipeline = pipeline("text2text-generation", model="google/flan-t5-base")

# Plant/Soil knowledge base
knowledge_base = {
    "best soil for gardening": "Loamy soil is best for gardening because it retains moisture and nutrients but also drains well.",
    "best soil for tomato": "Tomatoes grow best in well-drained loamy soil that is rich in organic matter.",
    "how to improve soil fertility": "You can improve soil fertility by adding compost, green manure, and crop rotation.",
    "ideal soil for strawberry": "Strawberries prefer slightly acidic, well-drained loamy soil with high organic content.",
    "what soil for roses": "Roses grow well in loamy, well-drained soil with a pH between 6.0 and 7.0."
}

# FastText-based language detection
def detect_language(text):
    lang = lang_model.predict(text)[0][0].replace("__label__", "")
    return lang

def generate_response(text):
    # Detect language and translate
    original_lang = detect_language(text)
    translated_input = GoogleTranslator(source='auto', target='en').translate(text)

    # Use knowledge base if match
    response_text = ""
    for key in knowledge_base:
        if key in translated_input.lower():
            response_text = knowledge_base[key]
            break

    # Otherwise use Hugging Face model
    if not response_text:
        response_text = qa_pipeline(translated_input, max_length=100)[0]['generated_text']

    # Translate back to original language
    translated_response = GoogleTranslator(source='en', target=original_lang).translate(response_text)

    # Text-to-speech
    try:
        tts = gTTS(text=translated_response, lang=original_lang)
    except:
        tts = gTTS(text=translated_response, lang='en')

    audio_path = os.path.join(tempfile.gettempdir(), f"{uuid.uuid4()}.mp3")
    tts.save(audio_path)

    return translated_response, audio_path

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("🌍 **Multilingual Voice Chatbot**\nAsk in any language about plants/soil!")

    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(label="🌱 Ask a Question")
            ask_button = gr.Button("🧠 Get Answer")

        with gr.Column():
            text_output = gr.Textbox(label="📝 Response")
            audio_output = gr.Audio(label="🔊 Voice", autoplay=True)

    ask_button.click(fn=generate_response, inputs=text_input, outputs=[text_output, audio_output])

demo.launch(share=True)