add instrumental stem

app.py

@@ -22,7 +22,7 @@ model_card = ModelCard(
 
 
 DEMUX_MODELS = ["mdx_extra_q", "mdx_extra", "htdemucs", "mdx_q"]
-STEM_NAMES = ["Drums", "Bass", "Other", "Vocals"]
+STEM_NAMES = ["Drums", "Bass", "Vocals", "Instrumental (No Vocals)"]
 
 # Global model cache
 LOADED_MODELS = {}
@@ -61,17 +61,22 @@ def separate_all_stems(audio_file_path: str, model_name: str):
         signal = AudioSignal(stem.cpu().numpy().astype("float32"), sample_rate=sr)
         output_signals.append(signal)
 
-    return output_signals  # [drums, bass, other, vocals]
+    # Combine drums + bass + other = instrumental
+    drums, bass, other, vocals = output_signals
+    instrumental = drums + bass + other
+
+    return [drums, bass, vocals, instrumental]
 
 # Process Function
-@spaces.GPU
 def process_fn(audio_file_path, model_name):
     output_signals = separate_all_stems(audio_file_path, model_name)
+
     outputs = []
     for stem_name, signal in zip(STEM_NAMES, output_signals):
-        filename = f"demucs_{model_name}_{stem_name.lower()}.wav"
+        filename = f"demucs_{model_name}_{stem_name.lower().replace(' ', '_')}.wav"
         output_audio_path = save_audio(signal, filename)
         outputs.append(output_audio_path)
+
     return tuple(outputs)
 
 # Gradio App 
@@ -87,9 +92,9 @@ with gr.Blocks() as demo:
     # Outputs: Multiple stems
     output_drums = gr.Audio(type="filepath", label="Drums")
     output_bass = gr.Audio(type="filepath", label="Bass")
-    output_other = gr.Audio(type="filepath", label="Other")
     output_vocals = gr.Audio(type="filepath", label="Vocals")
-    output_labels = gr.JSON(label="Labels")
+    output_instrumental = gr.Audio(type="filepath", label="Instrumental (No Vocals)")
+    #output_labels = gr.JSON(label="Labels")
 
     app = build_endpoint(
         model_card=model_card,
@@ -97,8 +102,8 @@ with gr.Blocks() as demo:
         output_components=[
             output_drums,
             output_bass,
-            output_other,
-            output_vocals
+            output_vocals,
+            output_instrumental
         ],
         process_fn=process_fn
     )