correct app.py

app.py

@@ -8,6 +8,9 @@ from typing import Dict
 from pyharp import *
 
 
+# ---------------------------
+# Config
+# ---------------------------
 
 DEMUX_MODELS = ["mdx_extra_q", "mdx_extra", "htdemucs", "mdx_q"]
 
@@ -19,8 +22,10 @@ STEM_CHOICES = {
     "Instrumental (No Vocals)": "instrumental"
 }
 
+# ---------------------------
+# Stem Separation Logic
+# ---------------------------
 
-# Stem Separation 
 def separate_stem(audio_file_path: str, model_name: str, stem_choice: str) -> AudioSignal:
     model = pretrained.get_model(model_name)
     model.to('cuda' if torch.cuda.is_available() else 'cpu')
@@ -53,78 +58,60 @@ def separate_stem(audio_file_path: str, model_name: str, stem_choice: str) -> Au
 
     return AudioSignal(stem.cpu().numpy().astype('float32'), sample_rate=sr)
 
-# Label & Metadata Handling
-def generate_dummy_metadata(stem_choice: str, username: str) -> Dict:
-    dummy_label = AudioLabel(
-        t=0.0,
-        label=stem_choice,
-        amplitude=0.7,
-        description=f"Start of {stem_choice} stem",
-        color=AudioLabel.hex_color_to_int("#FF5733")
-    )
-
-    label_list = LabelList(labels=[dummy_label])
-    label_list.meta["user"] = username
-
-    return {
-        "meta": label_list.meta,
-        "labels": [vars(label) for label in label_list.labels]
-    }
-
-import json
-
-def process_fn_stem(audio_file_path, demucs_model, stem_choice, profile=None):
-    username = profile.username if profile else "anonymous"
+# ---------------------------
+# Gradio Callback Function
+# ---------------------------
 
+def process_fn_stem(audio_file_path: str, demucs_model: str, stem_choice: str) -> Dict:
     stem_signal = separate_stem(audio_file_path, model_name=demucs_model, stem_choice=stem_choice)
     stem_filename = f"{stem_choice.lower().replace(' ', '_')}.wav"
     stem_path = save_audio(stem_signal, stem_filename)
 
-    metadata = {
-        "user": username,
+    return {
+        "file": stem_path,
         "stem": stem_choice,
-        "amplitude": 0.7,
-        "description": f"Start of {stem_choice}"
+        "description": f"Separated {stem_choice} stem",
+        "amplitude": 0.7
     }
 
-    return stem_path, json.dumps(metadata, indent=2)
-
+# ---------------------------
+# Model Card (for HARP)
+# ---------------------------
 
+model_card = ModelCard(
+    name="Demucs Stem Separator",
+    description="Uses Demucs to separate a music track into a selected stem.",
+    author="Alexandre Défossez, Nicolas Usunier, Léon Bottou, Francis Bach",
+    tags=["demucs", "source-separation", "pyharp", "stems"]
+)
 
+# ---------------------------
+# Gradio Interface (Blocks)
+# ---------------------------
 
-# Gradio Interface
 with gr.Blocks() as demo:
     gr.Markdown("# 🎧 Demucs Stem Separator")
     gr.Markdown("Log in with your Hugging Face account to separate music into stems.")
 
     gr.LoginButton()
 
-    with gr.Row():
-        model_dropdown = gr.Dropdown(
-            label="Select Demucs Model",
-            choices=DEMUX_MODELS,
-            value="mdx_extra_q"
-        )
-        stem_dropdown = gr.Dropdown(
-            label="Select Stem to Separate",
-            choices=list(STEM_CHOICES.keys()),
-            value="Vocals"
-        )
-
-    audio_input = gr.Audio(label="Upload Audio", type="filepath")
-
-    output_file = gr.File(label="Separated Stem (.wav)")
-    output_text = gr.Textbox(label="Metadata (Debug-safe JSON)", lines=6, max_lines=10)
-
-    run_button = gr.Button("Run Separation")
-
-    run_button.click(
-        fn=process_fn_stem,
-        inputs=[audio_input, model_dropdown, stem_dropdown],
-        outputs=[output_file, output_text]
+    dropdown_model = gr.Dropdown(
+        label="Select Demucs Model",
+        choices=DEMUX_MODELS,
+        value="mdx_extra_q"
     )
 
-    demo.queue()
+    dropdown_stem = gr.Dropdown(
+        label="Select Stem to Separate",
+        choices=list(STEM_CHOICES.keys()),
+        value="Vocals"
+    )
 
+    app = build_endpoint(
+        model_card=model_card,
+        components=[dropdown_model, dropdown_stem],
+        process_fn=process_fn_stem
+    )
 
+demo.queue()
 demo.launch(show_error=True)