import gradio as gr
import torch
import scipy.io.wavfile
import numpy as np
import tempfile
import os
from audiocraft.models import MusicGen, AudioGen

# Force l'utilisation du GPU si disponible
device = "cuda" if torch.cuda.is_available() else "cpu"

print(f"🔄 Chargement des modèles sur {device}...")
try:
    # On utilise 'small' pour la musique et 'medium' pour l'audio pour un bon ratio vitesse/qualité
    music_model = MusicGen.get_pretrained("facebook/musicgen-small")
    audio_model = AudioGen.get_pretrained("facebook/audiogen-medium")
except Exception as e:
    print(f"❌ Erreur lors du chargement des modèles : {e}")

def generate_music(prompt, duration=10):
    if not prompt: return None
    music_model.set_generation_params(duration=duration)
    wav = music_model.generate([prompt])
    wav_np = wav[0, 0].cpu().numpy()
    
    path = os.path.join(tempfile.gettempdir(), "music_out.wav")
    scipy.io.wavfile.write(path, music_model.sample_rate, (wav_np * 32767).astype(np.int16))
    return path

def generate_sound(prompt, duration=5):
    if not prompt: return None
    audio_model.set_generation_params(duration=duration)
    wav = audio_model.generate([prompt])
    wav_np = wav[0, 0].cpu().numpy()
    
    path = os.path.join(tempfile.gettempdir(), "sound_out.wav")
    scipy.io.wavfile.write(path, audio_model.sample_rate, (wav_np * 32767).astype(np.int16))
    return path

# Interface UI
with gr.Blocks(title="🎙️ Studio Audio IA", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🎙️ Studio Audio IA")
    gr.Markdown("Générez de la musique et des effets sonores avec les modèles de Facebook Research.")
    
    with gr.Tab("🎵 Musique (MusicGen)"):
        with gr.Row():
            with gr.Column():
                p1 = gr.Textbox(label="Description de la musique", placeholder="Ex: Epic cinematic orchestral music, high strings...")
                d1 = gr.Slider(minimum=1, maximum=30, value=10, step=1, label="Durée (secondes)")
                b1 = gr.Button("Générer la musique", variant="primary")
            with gr.Column():
                o1 = gr.Audio(label="Résultat Audio", type="filepath")
        b1.click(generate_music, inputs=[p1, d1], outputs=o1)

    with gr.Tab("🔊 Bruitages (AudioGen)"):
        with gr.Row():
            with gr.Column():
                p2 = gr.Textbox(label="Effet sonore", placeholder="Ex: Dogs barking, car engine revving, rain on window...")
                d2 = gr.Slider(minimum=1, maximum=15, value=5, step=1, label="Durée (secondes)")
                b2 = gr.Button("Générer le son", variant="primary")
            with gr.Column():
                o2 = gr.Audio(label="Résultat Audio", type="filepath")
        b2.click(generate_sound, inputs=[p2, d2], outputs=o2)

# Important : server_name="0.0.0.0" permet l'accès externe depuis Vast.ai
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, share=False)