handling mp3 input

app.py

@@ -1,6 +1,6 @@
 import spaces
 from pyharp.core import ModelCard, build_endpoint
-from pyharp.media.audio import load_audio, save_audio
+from pyharp.media.audio import save_audio
 from pyharp.labels import LabelList
 
 from audiotools import AudioSignal
@@ -8,7 +8,6 @@ from demucs import pretrained
 from demucs.apply import apply_model
 
 import gradio as gr
-import torchaudio
 import torch
 from pathlib import Path
 
@@ -20,7 +19,6 @@ model_card = ModelCard(
     tags=["demucs", "source-separation", "pyharp", "stems", "multi-output"]
 )
 
-
 DEMUX_MODELS = ["mdx_extra_q", "mdx_extra", "htdemucs", "mdx_q"]
 STEM_NAMES = ["Drums", "Bass", "Vocals", "Instrumental (No Vocals)"]
 
@@ -36,8 +34,10 @@ def get_cached_model(model_name: str):
         LOADED_MODELS[model_name] = model
     return LOADED_MODELS[model_name]
 
-# Separation Logic (all stems)
+# Separation Logic
 def separate_all_stems(audio_file_path: str, model_name: str):
+    model = get_cached_model(model_name)
+
     signal = AudioSignal(audio_file_path)
     signal = signal.resample(44100)  # expects 44.1kHz
 
@@ -46,19 +46,19 @@ def separate_all_stems(audio_file_path: str, model_name: str):
         signal = signal.convert_to(stereo=True)
 
     sr = signal.sample_rate  
+
     waveform = signal.audio_data.float()  # [channels, samples]
     waveform = waveform.unsqueeze(0)      # [1, channels, samples]
 
-
     with torch.no_grad():
         stems_batch = apply_model(
             model,
-            #waveform.unsqueeze(0).to(next(model.parameters()).device),
             waveform.to(next(model.parameters()).device),
             overlap=0.2,
             shifts=1,
             split=True,
         )
+
     stems = stems_batch[0]
 
     output_signals = []
@@ -86,13 +86,12 @@ def process_fn(audio_file_path, model_name):
         filename = f"demucs_{model_name}_{stem_name.lower().replace(' ', '_')}.{extension}"
         output_path = Path(filename)
 
-        # Use .export() to control output format
         signal.export(output_path, format=extension)
         outputs.append(str(output_path))
 
     return tuple(outputs)
 
-# Gradio App 
+# Gradio App
 with gr.Blocks() as demo:
     input_audio = gr.Audio(type="filepath", label="Input Audio").harp_required(True)
 
@@ -102,12 +101,10 @@ with gr.Blocks() as demo:
         value="htdemucs"
     )
 
-    # Outputs: Multiple stems
     output_drums = gr.Audio(type="filepath", label="Drums")
     output_bass = gr.Audio(type="filepath", label="Bass")
     output_vocals = gr.Audio(type="filepath", label="Vocals")
     output_instrumental = gr.Audio(type="filepath", label="Instrumental (No Vocals)")
-    #output_labels = gr.JSON(label="Labels")
 
     app = build_endpoint(
         model_card=model_card,