Update app.py

app.py

@@ -1,3 +1,87 @@
+# 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
+
+
+# from huggingface_hub import login
+
+# hf_token = os.getenv('HF_TOKEN')
+# login(token=hf_token,add_to_git_credential=True)
+
+# # Load the model outside of the GPU-decorated function
+# def load_model():
+#     print("Loading model...")
+#     model, model_config = get_pretrained_model("stabilityai/stable-audio-open-1.0")
+#     print("Model loaded successfully.")
+#     return model, model_config
+
+# # Define the function to generate audio
+# @spaces.GPU(duration=120)
+# def generate_audio(prompt, bpm, seconds_total):
+#     device = "cuda" if torch.cuda.is_available() else "cpu"
+
+#     # Download model
+#     model, model_config = load_model()
+#     sample_rate = model_config["sample_rate"]
+#     sample_size = model_config["sample_size"]
+
+#     model = model.to(device)
+
+#     # Set up text and timing conditioning
+#     conditioning = [{
+#         "prompt": f"{bpm} BPM {prompt}",
+#         "seconds_start": 0,
+#         "seconds_total": seconds_total
+#     }]
+
+#     # Generate stereo audio
+#     output = generate_diffusion_cond(
+#         model,
+#         steps=100,
+#         cfg_scale=7,
+#         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, and save to file
+#     output = output.to(torch.float32).div(torch.max(torch.abs(output))).clamp(-1, 1).mul(32767).to(torch.int16).cpu()
+    
+#     output_path = "output.wav"
+#     torchaudio.save(output_path, output, sample_rate)
+    
+#     return output_path
+
+# # Define the Gradio interface
+# iface = gr.Interface(
+#     fn=generate_audio,
+#     inputs=[
+#         gr.Textbox(label="Prompt", placeholder="Enter the description of the audio (e.g., tech house drum loop)"),
+#         gr.Number(label="BPM", value=128),
+#         gr.Number(label="Duration (seconds)", value=30)
+#     ],
+#     outputs=gr.Audio(label="Generated Audio"),
+#     title="Stable Audio Generation",
+#     description="Generate audio based on a text prompt using stable audio tools.",
+# )
+
+# # Launch the interface
+# iface.launch()
+
 import torch
 import torchaudio
 from einops import rearrange
@@ -10,12 +94,6 @@ import uuid
 from stable_audio_tools import get_pretrained_model
 from stable_audio_tools.inference.generation import generate_diffusion_cond
 
-
-from huggingface_hub import login
-
-hf_token = os.getenv('HF_TOKEN')
-login(token=hf_token,add_to_git_credential=True)
-
 # Load the model outside of the GPU-decorated function
 def load_model():
     print("Loading model...")
@@ -23,30 +101,43 @@ def load_model():
     print("Model loaded successfully.")
     return model, model_config
 
-# Define the function to generate audio
-@spaces.GPU(duration=120)
-def generate_audio(prompt, bpm, seconds_total):
+# 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):
+    print(f"Prompt received: {prompt}")
+    print(f"Settings: Duration={seconds_total}s, Steps={steps}, CFG Scale={cfg_scale}")
+
     device = "cuda" if torch.cuda.is_available() else "cpu"
+    print(f"Using device: {device}")
 
-    # Download model
+    # Fetch the Hugging Face token from the environment variable
+    hf_token = os.getenv('HF_TOKEN')
+    print(f"Hugging Face token: {hf_token}")
+
+    # Use pre-loaded model and configuration
     model, model_config = load_model()
     sample_rate = model_config["sample_rate"]
     sample_size = model_config["sample_size"]
 
+    print(f"Sample rate: {sample_rate}, Sample size: {sample_size}")
+
     model = model.to(device)
+    print("Model moved to device.")
 
     # Set up text and timing conditioning
     conditioning = [{
-        "prompt": f"{bpm} BPM {prompt}",
+        "prompt": prompt,
         "seconds_start": 0,
         "seconds_total": seconds_total
     }]
+    print(f"Conditioning: {conditioning}")
 
     # Generate stereo audio
+    print("Generating audio...")
     output = generate_diffusion_cond(
         model,
-        steps=100,
-        cfg_scale=7,
+        steps=steps,
+        cfg_scale=cfg_scale,
         conditioning=conditioning,
         sample_size=sample_size,
         sigma_min=0.3,
@@ -54,30 +145,44 @@ def generate_audio(prompt, bpm, seconds_total):
         sampler_type="dpmpp-3m-sde",
         device=device
     )
+    print("Audio generated.")
 
     # Rearrange audio batch to a single sequence
     output = rearrange(output, "b d n -> d (b n)")
+    print("Audio rearranged.")
 
-    # Peak normalize, clip, convert to int16, and save to file
+    # 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()
-    
-    output_path = "output.wav"
-    torchaudio.save(output_path, output, sample_rate)
-    
-    return output_path
+    print("Audio normalized and converted.")
+
+    # Generate a unique filename for the output
+    unique_filename = f"output_{uuid.uuid4().hex}.wav"
+    print(f"Saving audio to file: {unique_filename}")
+
+    # Save to file
+    torchaudio.save(unique_filename, output, sample_rate)
+    print(f"Audio saved: {unique_filename}")
 
-# Define the Gradio interface
-iface = gr.Interface(
+    # 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 the description of the audio (e.g., tech house drum loop)"),
-        gr.Number(label="BPM", value=128),
-        gr.Number(label="Duration (seconds)", value=30)
+        gr.Textbox(label="Prompt", placeholder="Enter your text prompt here"),
+        gr.Slider(0, 47, value=30, label="Duration in Seconds"),
+        gr.Slider(10, 150, value=100, step=10, label="Number of Diffusion Steps"),
+        gr.Slider(1, 15, value=7, step=0.1, label="CFG Scale")
     ],
-    outputs=gr.Audio(label="Generated Audio"),
-    title="Stable Audio Generation",
-    description="Generate audio based on a text prompt using stable audio tools.",
+    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
-iface.launch()
+# Pre-load the model to avoid multiprocessing issues
+model, model_config = load_model()
+
+# Launch the Interface
+interface.launch()
+