Update app.py

app.py

@@ -22,10 +22,14 @@ pipe = AudioLDMControlNetPipeline.from_pretrained(
 pipe = pipe.to(device)
 generator = torch.Generator(device)
 
-def predict(midi_file=None, prompt="", neg_prompt="", duration=10, seed=0, cond=1, inf=20, guidance_scale=2.5, guess=False):
+def predict(midi_file=None, midi_synth=None, prompt="", neg_prompt="", duration=None, seed=0, cond=1, inf=20, guidance_scale=2.5, guess=False):
     if isinstance(midi_file, _TemporaryFileWrapper):
         midi_file = midi_file.name
     midi = PrettyMIDI(midi_file)
+    if not duration or duration == 0:
+        print("Duration", duration)
+        duration = midi_synth[1].shape[0]/SAMPLE_RATE
+        print("New duration", duration)
     audio = pipe(
         prompt,
         negative_prompt=neg_prompt,
@@ -49,11 +53,7 @@ def synthesize(midi_file=None):
 
 def run_example(midi_file=None, prompt="", neg_prompt="", duration=None, seed=0, cond=1, inf=20, guidance_scale=2.5, guess=False):
     midi_synth = synthesize(midi_file)
-    if not duration or duration == 0:
-        print("Duration", duration)
-        duration = midi_synth[1].shape[0]/SAMPLE_RATE
-        print("New duration", duration)
-    gen_audio = predict(midi_file, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess)
+    gen_audio = predict(midi_file, midi_synth, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess)
     return midi_synth, gen_audio
 
 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:
@@ -83,7 +83,7 @@ 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=[audio])
+    btn.click(predict, inputs=[midi, midi_synth, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess], outputs=[audio])
     gr.Examples(examples=[["S00.mid", "piano", "", 10, 48, 1.0, 20, 2.5, False], ["S00.mid", "violin", "", 10, 48, 1.0, 20, 2.5, False], ["S00.mid", "woman singing, studio recording", "noise", 10, 48, 1.0, 20, 2.5, False], ["S00.mid", "jazz band, clean", "noise", 10, 48, 1.0, 20, 2.5, False], ["S00.mid", "choir", "noise, percussion", 10, 48, 1.0, 20, 2.5, False]], inputs=[midi, prompt, neg_prompt, duration, seed, cond, inf, guidance_scale, guess], fn=run_example, outputs=[midi_synth, audio], cache_examples=True)
     
 demo.launch()