app.py

import spaces
import torch
import torchaudio
from einops import rearrange
from stable_audio_tools import get_pretrained_model
from stable_audio_tools.inference.generation import generate_diffusion_cond
import gradio as gr
import os
from huggingface_hub import login

# Authentifizierung
if os.environ.get("HUGGING_FACE_HUB_TOKEN"):
    login(token=os.environ["HUGGING_FACE_HUB_TOKEN"])

@spaces.GPU(duration=180)
def generate_audio(prompt, duration=10, steps=50, cfg_scale=7):
    device = "cuda" if torch.cuda.is_available() else "cpu"
    
    # Modell laden und zum Gerät verschieben
    model, model_config = get_pretrained_model("stabilityai/stable-audio-open-1.0")
    model = model.to(device)
    
    sample_rate = model_config["sample_rate"]
    sample_size = model_config["sample_size"]

    # Konditionierung einrichten
    conditioning = [{
        "prompt": prompt,
        "seconds_start": 0,
        "seconds_total": duration
    }]

    # Audio generieren mit anpassbaren Parametern
    output = generate_diffusion_cond(
        model,
        steps=steps,
        cfg_scale=cfg_scale,
        conditioning=conditioning,
        sample_size=sample_size,
        sigma_min=0.3,
        sigma_max=500,
        sampler_type="dpmpp-3m-sde",  # Besserer Sampler
        device=device
    )

    # Audio-Batch in eine einzelne Sequenz umwandeln
    output = rearrange(output, "b d n -> d (b n)")

    # Peak-Normalisierung, Clipping, Konvertierung zu int16
    output = output.to(torch.float32).div(torch.max(torch.abs(output))).clamp(-1, 1).mul(32767).to(torch.int16).cpu()

    return output, sample_rate

def generate(prompt, duration=10, steps=50, cfg_scale=7):
    audio, sr = generate_audio(prompt, duration, steps, cfg_scale)
    return (sr, audio.numpy())

# Verbesserte Benutzeroberfläche
iface = gr.Interface(
    fn=generate,
    inputs=[
        gr.Textbox(label="Prompt", placeholder="Beschreiben Sie den gewünschten Sound..."),
        gr.Slider(minimum=1, maximum=30, value=10, step=1, label="Dauer (Sekunden)"),
        gr.Slider(minimum=20, maximum=100, value=50, step=5, label="Anzahl der Schritte"),
        gr.Slider(minimum=1, maximum=15, value=7, step=0.5, label="CFG Scale"),
    ],
    outputs=gr.Audio(label="Generiertes Audio"),
    title="Stable Audio Generator",
    description="Generieren Sie Audio aus Textbeschreibungen mit Stable Audio 1.0",
)

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