import gradio as gr
import spaces
import os, torch, io
import json
import re
# os.system("python -m unidic download")
import httpx
# print("Make sure you've downloaded unidic (python -m unidic download) for this WebUI to work.")
from melo.api import TTS
import tempfile
import wave
from pydub import AudioSegment
from gradio_client import Client

client = Client("eswardivi/AIO_Chat")
def fetch_text(url):
    print("Entered Webpage Extraction")
    prefix_url = "https://r.jina.ai/"
    url = prefix_url + url
    response = httpx.get(url, timeout=120.0)
    print("Response Received")
    return response.text


@spaces.GPU
def synthesize(article_url, progress=gr.Progress()):
    text = fetch_text(article_url)
    device = "cuda" if torch.cuda.is_available() else "cpu"
    
    template = """
        {
            "conversation": [
                {"speaker": "", "text": ""},
                {"speaker": "", "text": ""}
            ]
        }
        """

    result = client.predict(
		f"{text} \n Convert the text as Elaborate Conversation between two people as Podcast.\nfollowing this template and return only JSON \n {template}",
		0.9,	
		True,	
		1024,
		api_name="/chat"
    )
    pattern = r"\{(?:[^{}]|(?:\{[^{}]*\}))*\}"
    json_match = re.search(pattern, result)
    if json_match:
        conversation=json_match.group()
    else:
        conversation = template
    speed = 1.0
    models = {
        "EN": TTS(language="EN", device=device),
    }
    speakers = ["EN-Default", "EN-US"]

    combined_audio = AudioSegment.empty()
    conversation = json.loads(conversation)
    for i, turn in enumerate(conversation["conversation"]):
        bio = io.BytesIO()
        text = turn["text"]
        speaker = speakers[i % 2]
        speaker_id = models["EN"].hps.data.spk2id[speaker]
        models["EN"].tts_to_file(
            text, speaker_id, bio, speed=speed, pbar=progress.tqdm, format="wav"
        )
        bio.seek(0)
        audio_segment = AudioSegment.from_file(bio, format="wav")
        combined_audio += audio_segment

    final_audio_path = "final.mp3"
    combined_audio.export(final_audio_path, format="mp3")
    return final_audio_path


with gr.Blocks() as demo:
    gr.Markdown("# Not Ready to USE")
    gr.Markdown("# Turn Any Article into Podcast")
    gr.Markdown("## Easily convert articles from URLs into listenable audio Podcast.")
    with gr.Group():
        text = gr.Textbox(label="Article Link")
    btn = gr.Button("Podcasitfy", variant="primary")
    aud = gr.Audio(interactive=False)
    btn.click(synthesize, inputs=[text], outputs=[aud])

demo.queue(api_open=True, default_concurrency_limit=10).launch(show_api=True,share=True)