Add score as image

app.py

@@ -8,6 +8,7 @@ import gradio as gr
 import note_seq
 from pytube import YouTube
 from pydub import AudioSegment
+from music21 import converter
 
 from inferencemodel import InferenceModel
 from utils import upload_audio, create_image_from_note_sequence
@@ -72,11 +73,15 @@ def inference(yt_audio_path):
     int16_data = note_seq.audio_io.float_samples_to_int16(array_of_floats)
     piano_roll = create_image_from_note_sequence(est_ns)
     
-    return "./transcribed.mid", (SAMPLE_RATE, int16_data), piano_roll
+    parsed = converter.parse("./transcribed.mid")
+    parsed.write("./transcribed.png")
+    
+    return "./transcribed.mid", (SAMPLE_RATE, int16_data), piano_roll, "./transcribed.png"
   
 title = "Transcribe music from YouTube videos using Transformers."
 description = """
 Gradio demo for Music Transcription with Transformers. Read more in the links below.
+To use this demo, just add a YouTube link with the music you want to transcribe.
 """
 article = "<p style='text-align: center'><a href='https://magenta.tensorflow.org/transcription-with-transformers' target='_blank'>Blog: Music Transcription with Transformers</a> | <a href='https://github.com/magenta/mt3' target='_blank'>Github Repo</a></p>"
 
@@ -120,10 +125,11 @@ with demo:
             midi_audio = gr.Audio()
         with gr.Row():
             piano_roll = gr.Image()
+            score = gr.Image()
         
         btn.click(inference,
                   inputs=yt_audio_path,
-                  outputs=[midi_file, midi_audio, piano_roll])
+                  outputs=[midi_file, midi_audio, piano_roll, score])
         
         gr.Markdown(article)