import gradio as gr
from langchain.prompts import PromptTemplate
from langchain_huggingface import HuggingFacePipeline
from transformers import pipeline
from bs4 import BeautifulSoup
import requests
from TTS.api import TTS
import tempfile
import os

# Setup summarization LLM
summary_pipe = pipeline("text2text-generation", model="google/flan-t5-base", device=-1)
llm = HuggingFacePipeline(pipeline=summary_pipe)

# Prompt for more engaging summary
summary_prompt = PromptTemplate.from_template("""
Summarize the following article content in a clear, warm, and motivational tone like a preacher speaking to an audience:

{text}

Summary:
""")

summary_chain = summary_prompt | llm

# TTS model setup
tts_model = TTS(model_name="tts_models/multilingual/multi-dataset/your_tts", progress_bar=False, gpu=False)
SPEAKER_LIST = tts_model.speakers
DEFAULT_SPEAKER = SPEAKER_LIST[0] if SPEAKER_LIST else None

def extract_main_content(url):
    try:
        response = requests.get(url, timeout=10)
        soup = BeautifulSoup(response.content, "html.parser")
        for tag in soup(["nav", "header", "footer", "aside", "script", "style", "noscript"]):
            tag.decompose()
        paragraphs = soup.find_all("p")
        content = "\n".join([p.get_text() for p in paragraphs if len(p.get_text()) > 60])
        return content.strip()
    except Exception as e:
        return f"Error extracting article content: {str(e)}"

def generate_human_like_audio(text, speaker):
    try:
        # Save in tmp directory accessible by Gradio
        base = tempfile.gettempdir()
        wav_path = os.path.join(base, "summary.wav")
        mp3_path = os.path.join(base, "summary.mp3")

        tts_model.tts_to_file(text=text, speaker=speaker, file_path=wav_path)
        os.system(f"ffmpeg -y -i {wav_path} -codec:a libmp3lame -qscale:a 4 {mp3_path}")

        return wav_path if os.path.exists(wav_path) else None, mp3_path if os.path.exists(mp3_path) else None
    except Exception as e:
        print(f"TTS ERROR: {e}")
        return None, None

def url_to_audio_summary(url, speaker):
    article_text = extract_main_content(url)
    if article_text.startswith("Error"):
        return article_text, None, None

    # Truncate
    if len(article_text) > 1500:
        article_text = article_text[:1500] + "..."

    summary = (summary_chain.invoke if hasattr(summary_chain, 'invoke') else summary_chain)({"text": article_text})
    summary = summary.get("text") if isinstance(summary, dict) else summary

    wav_path, mp3_path = generate_human_like_audio(summary, speaker)
    return summary, wav_path, mp3_path

def interface_wrapper(url, speaker):
    summary, wav_path, mp3_path = url_to_audio_summary(url, speaker)
    return summary, wav_path, mp3_path

iface = gr.Interface(
    fn=interface_wrapper,
    inputs=[
        gr.Textbox(label="Article URL", placeholder="Paste a news/blog URL here..."),
        gr.Dropdown(choices=SPEAKER_LIST, value=DEFAULT_SPEAKER, label="Select Speaker")
    ],
    outputs=[
        gr.Textbox(label="Summary"),
        gr.Audio(label="Preacher-style Audio Summary", type="filepath"),
        gr.File(label="Download MP3")
    ],
    title="Preaching-Style URL to Audio Agent",
    description="Summarizes article content and reads it aloud in a warm, preacher-style voice using YourTTS. CPU-only."
)

if __name__ == "__main__":
    iface.launch()