File size: 2,667 Bytes
456ed62
 
 
 
 
a17cbc4
 
456ed62
 
 
 
 
 
 
 
 
a17cbc4
 
 
 
 
456ed62
 
a17cbc4
456ed62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a17cbc4
 
 
456ed62
a17cbc4
456ed62
 
a17cbc4
456ed62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
import torch
import torchaudio
from einops import rearrange
import gradio as gr
import spaces
import os
import uuid

# Importing the model-related functions
from stable_audio_tools import get_pretrained_model
from stable_audio_tools.inference.generation import generate_diffusion_cond

# Function to set up, generate, and process the audio
@spaces.GPU(duration=120)  # Allocate GPU only when this function is called
def generate_audio(prompt, seconds_total=30, steps=100, cfg_scale=7):
    device = "cuda" if torch.cuda.is_available() else "cpu"

    # Fetch the Hugging Face token from the environment variable
    hf_token = os.getenv('HF_TOKEN')
    if not hf_token:
        raise EnvironmentError("HF_TOKEN environment variable not set")
    
    # Download and set up the model
    model, model_config = get_pretrained_model("stabilityai/stable-audio-open-1.0", use_auth_token=hf_token)
    sample_rate = model_config["sample_rate"]
    sample_size = model_config["sample_size"]

    model = model.to(device)

    # Set up text and timing conditioning
    conditioning = [{
        "prompt": prompt,
        "seconds_start": 0,
        "seconds_total": seconds_total
    }]

    # Generate stereo audio
    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",
        device=device
    )

    # Rearrange audio batch to a single sequence
    output = rearrange(output, "b d n -> d (b n)")

    # Peak normalize, clip, convert to int16
    output = output.to(torch.float32).div(torch.max(torch.abs(output))).clamp(-1, 1).mul(32767).to(torch.int16).cpu()

    # Generate a unique filename for the output
    unique_filename = f"output_{uuid.uuid4().hex}.wav"
    # Save to file
    torchaudio.save(unique_filename, output, sample_rate)

    # Return the path to the generated audio file
    return unique_filename

# Setting up the Gradio Interface
interface = gr.Interface(
    fn=generate_audio,
    inputs=[
        gr.Textbox(label="Prompt", placeholder="Enter your text prompt here"),
        gr.Slider(0, 47, value=30, label="Duration in Seconds"),
        gr.Slider(10, 300, value=100, step=10, label="Number of Diffusion Steps"),
        gr.Slider(1, 15, value=7, step=0.1, label="CFG Scale")
    ],
    outputs=gr.Audio(type="filepath", label="Generated Audio"),
    title="Stable Audio Generator",
    description="Generate variable-length stereo audio at 44.1kHz from text prompts using Stable Audio Open 1.0."
)

# Launch the Interface
interface.launch()