Update app.py

app.py

@@ -26,7 +26,10 @@ generator = torch.Generator(device)
 def predict(midi_file=None, prompt="", negative_prompt="", audio_length_in_s=10, random_seed=0, controlnet_conditioning_scale=1, num_inference_steps=20, guidance_scale=2.5, guess_mode=False):
  if isinstance(midi_file, _TemporaryFileWrapper):
  midi_file = midi_file.name
- midi = PrettyMIDI(midi_file)
+ if not isinstance(midi_file, PrettyMIDI):
+ midi = PrettyMIDI(midi_file)
+ else:
+ midi = midi_file
  audio = pipe(
  prompt,
  negative_prompt=negative_prompt,
@@ -46,7 +49,7 @@ def synthesize(midi_file=None):
  midi = PrettyMIDI(midi_file)
  midi_synth = midi.synthesize(fs=SAMPLE_RATE)
  midi_synth = midi_synth.reshape(midi_synth.shape[0], 1)
- return (SAMPLE_RATE, midi_synth)
+ return midi, (SAMPLE_RATE, midi_synth)
 
 with gr.Blocks(title="🎹 MIDI-AudioLDM", theme=gr.themes.Base(text_size=gr.themes.sizes.text_md, font=[gr.themes.GoogleFont("Nunito Sans")])) as demo:
  gr.HTML(
@@ -60,8 +63,9 @@ with gr.Blocks(title="🎹 MIDI-AudioLDM", theme=gr.themes.Base(text_size=gr.the
  with gr.Row():
  with gr.Column(variant='panel'):
  midi_synth = gr.Audio(label="synthesized midi")
- midi = gr.UploadButton("Upload a MIDI File", file_types=[".mid"], value="S00.mid")
- midi.upload(synthesize, midi, midi_synth)
+ midi_file = gr.UploadButton("Upload a MIDI File", file_types=[".mid"], value="S00.mid")
+ midi = PrettyMIDI()
+ midi_file.upload(synthesize, midi_file, [midi, midi_synth])
  prompt = gr.Textbox(label="prompt", info="Enter a descriptive text prompt to guide the audio generation.")
  with gr.Column(variant='panel'):
  audio = gr.Audio(label="generated audio")