app.py created.
Browse files
app.py
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import spaces
|
| 2 |
+
import torch
|
| 3 |
+
import gradio as gr
|
| 4 |
+
from diffusers import StableAudioPipeline
|
| 5 |
+
|
| 6 |
+
# Initialize pipeline
|
| 7 |
+
pipe = StableAudioPipeline.from_pretrained(
|
| 8 |
+
"stabilityai/stable-audio-open-1.0",
|
| 9 |
+
torch_dtype=torch.float16
|
| 10 |
+
)
|
| 11 |
+
|
| 12 |
+
@spaces.GPU(duration=60) # Allocates GPU dynamically when run
|
| 13 |
+
def generate_audio(prompt, negative_prompt, seconds):
|
| 14 |
+
pipe.to("cuda")
|
| 15 |
+
output = pipe(
|
| 16 |
+
prompt=prompt,
|
| 17 |
+
negative_prompt=negative_prompt,
|
| 18 |
+
audio_end_in_s=seconds,
|
| 19 |
+
num_inference_steps=100,
|
| 20 |
+
)
|
| 21 |
+
return (44100, output.audios[0].T.cpu().numpy())
|
| 22 |
+
|
| 23 |
+
# Gradio Interface setup
|
| 24 |
+
demo = gr.Interface(
|
| 25 |
+
fn=generate_audio,
|
| 26 |
+
inputs=[
|
| 27 |
+
gr.Textbox(label="Audio Prompt"),
|
| 28 |
+
gr.Textbox(label="Negative Prompt", value="low quality, noise, distortion"),
|
| 29 |
+
gr.Slider(minimum=5, maximum=47, value=15, label="Duration (Seconds)")
|
| 30 |
+
],
|
| 31 |
+
outputs=gr.Audio(label="Generated SFX"),
|
| 32 |
+
title="My Custom Stable Audio Studio"
|
| 33 |
+
)
|
| 34 |
+
|
| 35 |
+
demo.launch()
|