Change audio type to filepath

app.py

@@ -64,9 +64,9 @@ def inference(yt_audio):
     est_ns = inference_model(audio)
     
     note_seq.sequence_proto_to_midi_file(est_ns, "./transcribed.mid")
-    note_sequence = note_seq.midi_to_note_sequence("./transcribed.mid")
+
     synth = note_seq.midi_synth.fluidsynth
-    array_of_floats = synth(note_sequence, sample_rate=44100)
+    array_of_floats = synth(est_ns, sample_rate=44100)
     int16_data = note_seq.audio_io.float_samples_to_int16(array_of_floats)
     # piano_roll = create_image_from_note_sequence(note_sequence)
     
@@ -105,7 +105,7 @@ with demo:
             title = gr.Label(label="Video Title", placeholder="Title")
             img = gr.Image(label="Thumbnail")
         with gr.Row():
-            yt_audio = gr.Audio()
+            yt_audio = gr.Audio(type="filepath", label="First 10 seconds")
 
         link.change(fn=populate_metadata, inputs=link, outputs=[img, title, yt_audio])
         

inferencemodel.py

@@ -18,9 +18,6 @@ from mt3 import preprocessors
 from mt3 import spectrograms
 from mt3 import vocabularies
 
-import nest_asyncio
-nest_asyncio.apply()
-
 class InferenceModel(object):
   """Wrapper of T5X model for music transcription."""