Return audio and audio path

app.py

@@ -53,11 +53,14 @@ def get_audio(url):
 def populate_metadata(link):
     yt = YouTube(link)
     audio = get_audio(link)
-    return yt.thumbnail_url, yt.title, audio
+    return yt.thumbnail_url, yt.title, audio, audio
 
-def inference(yt_audio):
-
-    audio = upload_audio(yt_audio,sample_rate=SAMPLE_RATE)
+def inference(yt_audio_path):
+    
+    with open(yt_audio_path, 'rb') as fd:
+      contents = fd.read()
+    
+    audio = upload_audio(contents,sample_rate=SAMPLE_RATE)
 
     est_ns = inference_model(audio)
     
@@ -104,8 +107,9 @@ with demo:
             img = gr.Image(label="Thumbnail")
         with gr.Row():
             yt_audio = gr.Audio()
+            yt_audio_path = gr.Textbox(visible=False)
 
-        link.change(fn=populate_metadata, inputs=link, outputs=[img, title, yt_audio])
+        link.change(fn=populate_metadata, inputs=link, outputs=[img, title, yt_audio, yt_audio_path])
         
         with gr.Row():
             btn = gr.Button("Transcribe music")
@@ -114,7 +118,8 @@ with demo:
             midi_file = gr.File()
             midi_audio = gr.Audio()
         
-        btn.click(inference(yt_audio),
+        btn.click(inference,
+                  inputs=yt_audio_path,
                   outputs=[midi_file, midi_audio])
         
         gr.Markdown(article)