app.py

import gradio as gr
import os
import asyncio
import json
from conver import ConversationConfig, URLToAudioConverter
from dotenv import load_dotenv
from pydub import AudioSegment

load_dotenv()

MUSICA_FONDO = "musica.mp3"
TAG1 = "tag.mp3"
TAG2 = "tag2.mp3"

def mezclar_musica_y_tags(audio_path: str, custom_music_path: str = None) -> str:
    podcast_audio = AudioSegment.from_file(audio_path)
    music_file = custom_music_path if custom_music_path and os.path.exists(custom_music_path) else MUSICA_FONDO
    musica_fondo = AudioSegment.from_file(music_file).apply_gain(-15)
    tag_outro = AudioSegment.from_file(TAG1).apply_gain(-5)
    tag_trans = AudioSegment.from_file(TAG2).apply_gain(-5)

    duracion_podcast = len(podcast_audio)
    repeticiones = (duracion_podcast // len(musica_fondo)) + 1
    musica_fondo_loop = musica_fondo * repeticiones
    musica_fondo_loop = musica_fondo_loop[:duracion_podcast]

    mezcla = musica_fondo_loop.overlay(podcast_audio)
    mezcla = mezcla + tag_outro

    silent_ranges = []
    for i in range(0, len(podcast_audio) - 500, 100):
        chunk = podcast_audio[i:i+500]
        if chunk.dBFS < -40:
            silent_ranges.append((i, i + 500))
    for start, end in reversed(silent_ranges):
        if (end - start) >= len(tag_trans):
            mezcla = mezcla.overlay(tag_trans, position=start + 50)

    output_path = audio_path.replace(".mp3", "_con_musica.mp3")
    mezcla.export(output_path, format="mp3")
    return output_path

async def generate_dialogue(article_url, text_input, language, skip_llm, custom_prompt):
    if not article_url and not text_input:
        return "Error: Ingresa una URL o texto", None

    try:
        config = ConversationConfig(custom_prompt_template=custom_prompt)
        converter = URLToAudioConverter(config, llm_api_key=os.environ.get("TOGETHER_API_KEY"))

        if skip_llm and text_input:
            dialogue = {"conversation": [{"speaker": "Anfitrión1", "text": text_input}]}
        elif text_input:
            dialogue = converter.extract_conversation(text_input)
        else:
            dialogue = converter.extract_conversation(await converter.fetch_text(article_url))

        # Usar comillas dobles al mostrar el JSON
        return json.dumps(dialogue, indent=2, ensure_ascii=False), dialogue
    except Exception as e:
        return f"Error: {str(e)}", None

async def generate_audio(dialogue_json, language, agregar_musica, custom_music):
    try:
        # Validar JSON antes de parsear
        json_str = dialogue_json.strip()
        try:
            dialogue = json.loads(json_str)
        except json.JSONDecodeError as e:
            return f"Error: JSON inválido - {str(e)}", None

        if not dialogue.get("conversation"):
            return "Error: El JSON no contiene 'conversation'", None

        config = ConversationConfig()
        converter = URLToAudioConverter(config, llm_api_key=os.environ.get("TOGETHER_API_KEY"))
        
        voices = {
            "en": ("en-US-AvaMultilingualNeural", "en-US-AndrewMultilingualNeural"),
            "es": ("es-ES-AlvaroNeural", "es-ES-ElviraNeural")
        }
        voice1, voice2 = voices.get(language, voices["en"])

        output_file, conversation = await converter._process_to_audio(dialogue, voice1, voice2, custom_music)

        if agregar_musica:
            output_file = mezclar_musica_y_tags(output_file, custom_music)

        return conversation, output_file
    except Exception as e:
        return f"Error: {str(e)}", None

def synthesize_sync(article_url, text_input, language, skip_llm, custom_prompt):
    return asyncio.run(generate_dialogue(article_url, text_input, language, skip_llm, custom_prompt))

def generate_audio_sync(dialogue_json, language, agregar_musica, custom_music):
    return asyncio.run(generate_audio(dialogue_json, language, agregar_musica, custom_music))

with gr.Blocks(theme='gstaff/sketch') as demo:
    gr.Markdown("# 🎙 Podcast Converter")
    with gr.Group():
        text_url = gr.Textbox(label="URL (opcional)", placeholder="https://...")
        text_input = gr.Textbox(label="Texto manual", lines=5, placeholder="Pega tu texto aquí...")
        language = gr.Dropdown(["en", "es"], label="Idioma", value="en")
        skip_llm = gr.Checkbox(label="🔴 Modo libre (sin filtros LLM)", value=False)
        custom_prompt = gr.Textbox(
            label="Prompt personalizado (opcional)",
            placeholder='{text}\nCrea un diálogo de podcast en español entre Anfitrión1 y Anfitrión2. Usa un tono informal y genera al menos 6 intercambios por hablante. Devuelve SOLO un objeto JSON: {"conversation": [{"speaker": "Anfitrión1", "text": "..."}, {"speaker": "Anfitrión2", "text": "..."}]}'
        )
        btn_dialogue = gr.Button("Generar Diálogo", variant="primary")
    
    with gr.Group():
        dialogue_json = gr.Textbox(label="Diálogo JSON (editable)", lines=10, interactive=True)
        agregar_musica = gr.Checkbox(label="🎵 Agregar música de fondo y cortinillas", value=False)
        custom_music = gr.File(label="Subir música de fondo (opcional)", file_types=[".mp3"])
        btn_audio = gr.Button("Generar Audio", variant="primary")
    
    with gr.Row():
        conv_display = gr.Textbox(label="Conversación", interactive=False, lines=10)
        aud = gr.Audio(label="Audio Generado", interactive=False)
    
    btn_dialogue.click(
        synthesize_sync,
        inputs=[text_url, text_input, language, skip_llm, custom_prompt],
        outputs=[dialogue_json, dialogue_json]
    )
    btn_audio.click(
        generate_audio_sync,
        inputs=[dialogue_json, language, agregar_musica, custom_music],
        outputs=[conv_display, aud]
    )

demo.launch()