Update app.py

app.py

@@ -27,8 +27,6 @@ def predict(midi_file=None, prompt="", negative_prompt="", audio_length_in_s=10,
     if isinstance(midi_file, _TemporaryFileWrapper):
         midi_file = midi_file.name
     midi = PrettyMIDI(midi_file)
-    midi_synth = midi.synthesize(fs=SAMPLE_RATE)[:int(SAMPLE_RATE*audio_length_in_s)]
-    midi_synth = midi_synth.reshape(midi_synth.shape[0], 1)
     audio = pipe(
         prompt,
         negative_prompt=negative_prompt,
@@ -40,8 +38,15 @@ def predict(midi_file=None, prompt="", negative_prompt="", audio_length_in_s=10,
         generator=generator.manual_seed(int(random_seed)),
         guidance_scale=float(guidance_scale),
     )
-    return (SAMPLE_RATE, midi_synth), (SAMPLE_RATE, audio.audios.T)
-    
+    return (SAMPLE_RATE, audio.audios.T)
+
+def synthesize(midi_file=None):
+    if isinstance(midi_file, _TemporaryFileWrapper):
+        midi_file = midi_file.name
+    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)
 
 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(
@@ -55,9 +60,10 @@ with gr.Blocks(title="🎹 MIDI-AudioLDM", theme=gr.themes.Base(text_size=gr.the
     with gr.Row():
         with gr.Column(variant='panel'):
             midi = gr.File(label="midi file", file_types=[".mid"])
+            midi_synth = gr.Audio(label="synthesized midi")
+            midi.upload(synthesize, inputs=midi, outputs=midi_synth)
             prompt = gr.Textbox(label="prompt", info="Enter a descriptive text prompt to guide the audio generation.")
         with gr.Column(variant='panel'):
-            synth = gr.Audio(label="synthesized audio")
             audio = gr.Audio(label="generated audio")
     with gr.Accordion("Advanced Settings", open=False):
         duration = gr.Slider(0, 30, value=10, step=2.5, label="duration", info="Modify the duration in seconds of the output audio file.")
@@ -68,7 +74,7 @@ with gr.Blocks(title="🎹 MIDI-AudioLDM", theme=gr.themes.Base(text_size=gr.the
         cond = gr.Slider(0.0, 1.0, value=1.0, step=0.1, label="conditioning scale", info="Choose a value between 0 and 1. The larger the more it will take the conditioning into account. Lower values are recommended for more creative prompts.")
         guess = gr.Checkbox(label="guess mode", info="Optionally select guess mode. If so, the model will try to recognize the content of the MIDI without the need of a text prompt.")
     btn = gr.Button("Generate")
-    btn.click(predict, inputs=[midi, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess], outputs=[synth, audio])
-    gr.Examples(examples=[["S00.mid", "piano", "", 10, 25, 1.0, 20, 2.5, False], ["S00.mid", "violin", "", 10, 25, 1.0, 20, 2.5, False], ["S00.mid", "woman singing, studio recording", "noise", 10, 25, 1.0, 20, 2.5, False], ["S00.mid", "jazz band, clean", "noise", 10, 25, 0.8, 20, 2.5, False], ["S00.mid", "choir", "noise, percussion", 10, 25, 0.7, 20, 2.5, False]], inputs=[midi, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess], fn=predict, outputs=[synth, audio], cache_examples=True)
+    btn.click(predict, inputs=[midi, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess], outputs=[audio])
+    gr.Examples(examples=[["S00.mid", "piano", "", 10, 25, 1.0, 20, 2.5, False], ["S00.mid", "violin", "", 10, 25, 1.0, 20, 2.5, False], ["S00.mid", "woman singing, studio recording", "noise", 10, 25, 1.0, 20, 2.5, False], ["S00.mid", "jazz band, clean", "noise", 10, 25, 0.8, 20, 2.5, False], ["S00.mid", "choir", "noise, percussion", 10, 25, 0.7, 20, 2.5, False]], inputs=[midi, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess], fn=predict, outputs=[audio], cache_examples=True)
     
 demo.launch()