Update app.py

app.py

@@ -12,7 +12,7 @@ translator = Translator()
 
 # Constants
 model = "stabilityai/stable-audio-open-1.0"
-# MAX_SEED = np.iinfo(np.int32).max
+MAX_SEED = np.iinfo(np.int32).max
 
 CSS = """
 .gradio-container {
@@ -41,8 +41,8 @@ a T5-based text embedding for text conditioning, and a transformer-based diffusi
 if torch.cuda.is_available():
     pipe = StableAudioPipeline.from_pretrained(
         model,
-        low_cpu_mem_usage=True,
-        torch_dtype=torch.float16).to("cuda")
+        torch_dtype=torch.float16)
+    pipe = pipe.to("cuda")
 
 
 # Function 
@@ -52,10 +52,10 @@ def main(
     negative="low quality",
     second: float = 10.0):
     
-    # if seed == -1:
-    #     seed = random.randint(0, MAX_SEED)
-    # seed = int(seed)
-    # generator = torch.Generator().manual_seed(seed)
+    if seed == -1:
+         seed = random.randint(0, MAX_SEED)
+    seed = int(seed)
+    generator = torch.Generator().manual_seed(seed)
     
     prompt = str(translator.translate(prompt, 'English'))
 
@@ -65,6 +65,9 @@ def main(
         prompt, 
         negative_prompt=negative, 
         audio_end_in_s=second,
+        num_inference_steps=200,
+        num_waveforms_per_prompt=3,
+        generator=generator,
     ).audios
 
     os.makedirs("outputs", exist_ok=True)
@@ -73,26 +76,25 @@ def main(
     
     sf.write(audio_path, audio[0].T.float().cpu().numpy(), pipe.vae.samping_rate)
     
-    return audio_path
+    return audio_path, seed
 
 # Gradio Interface
 
 with gr.Blocks(theme='soft', css=CSS, js=JS, title="Stable Audio Open") as iface:
     with gr.Accordion(""):
         gr.Markdown(DESCRIPTION)
+    output = gr.Audio(label="Podcast", type="filepath", interactive=False, autoplay=True, elem_classes="audio")  # Create an output textbox
+    prompt = gr.Textbox(label="Prompt", placeholder="1000 BPM percussive sound of water drops")
+    negative = gr.Textbox(label="Negative prompt", placeholder="Low quality")
     with gr.Row():
-        output = gr.Audio(label="Podcast", type="filepath", interactive=False, autoplay=True, elem_classes="audio")  # Create an output textbox
-    with gr.Row():
-        prompt = gr.Textbox(label="Prompt", placeholder="1000 BPM percussive sound of water drops")
-    with gr.Row():
-        negative = gr.Textbox(label="Negative prompt", placeholder="Low quality")
         second =  gr.Slider(5.0, 60.0, value=10.0, label="Second", step=0.1),
+        seed = gr.Slider(1, MAX_SEED, value=0, label="Seed", step=1),
     with gr.Row():
         submit_btn = gr.Button("🚀 Send")  # Create a submit button
-        clear_btn = gr.ClearButton(output, value="🗑️ Clear") # Create a clear button
+        clear_btn = gr.ClearButton([prompt, seed, output], value="🗑️ Clear") # Create a clear button
 
     # Set up the event listeners
-    submit_btn.click(main, inputs=[prompt, negative, second], outputs=output)
+    submit_btn.click(main, inputs=[prompt, negative, second, seed], outputs=[output, seed])
 
 
 #gr.close_all()