Get metadata from youtube link

app.py

@@ -5,6 +5,7 @@ os.system("python3 -m pip install -e .")
 import gradio as gr
 
 import note_seq
+from pytube import YouTube
 
 from inferencemodel import InferenceModel
 from utils import upload_audio
@@ -24,6 +25,11 @@ def change_model(model):
     inference_model = InferenceModel("/home/user/app/checkpoints/mt3/", model)
     current_model = model
 
+# Credits https://huggingface.co/spaces/jeffistyping/Youtube-Whisperer
+def populate_metadata(link):
+    yt = YouTube(link)
+    return yt.thumbnail_url, yt.title
+
 def inference(audio):
     with open(audio, "rb") as fd:
         contents = fd.read()
@@ -37,8 +43,8 @@ def inference(audio):
     return "./transcribed.mid"
   
 title = "Transcribe music from YouTube videos using Transformers."
-description = """"
-Gradio demo for Music Transcription with Transformers Read more in the links below.
+description = """
+Gradio demo for Music Transcription with Transformers. Read more in the links below.
 """
 
 article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2111.03017' target='_blank'>MT3: Multi-Task Multitrack Music Transcription</a> | <a href='https://github.com/magenta/mt3' target='_blank'>Github Repo</a></p>"
@@ -53,16 +59,22 @@ with demo:
                 + "</h1>")
     gr.Markdown(description)
     with gr.Box():
-        with gr.Row():
-            gr.Markdown("<h2>Select your model</h2>")
-            gr.Markdown("""
-                        The ismir2021 model transcribes piano only, with note velocities. 
-                        The mt3 model transcribes multiple simultaneous instruments, but without velocities."
-                        """)
-            model = gr.Radio(
-                ["mt3", "ismir2021"], label="What kind of model you want to use?"
-            )
-            model.change(fn=change_model, inputs=model, outputs=[])
+        gr.Markdown("<h2>Select your model</h2>")
+        gr.Markdown("""
+                    The ismir2021 model transcribes piano only, with note velocities. 
+                    The mt3 model transcribes multiple simultaneous instruments, but without velocities."
+                    """)
+        model = gr.Radio(
+            ["mt3", "ismir2021"], label="What kind of model you want to use?"
+        )
+        model.change(fn=change_model, inputs=model, outputs=[])
+
+        link = gr.Textbox(label="YouTube Link")
+        with gr.Row().style(mobile_collapse=False, equal_height=True):
+            title = gr.Label(label="Video Title", placeholder="Title")
+            img = gr.Image(label="Thumbnail")
+
+        link.change(fn=populate_metadata, inputs=link, outputs=[img, title])
 
 demo.launch()  
   

requirements.txt

@@ -7,4 +7,5 @@ jax[cpu]==0.3.15 -f https://storage.googleapis.com/jax-releases/jax_releases.htm
 # pin CLU for python 3.7 compatibility
 clu==0.0.7
 # pin Orbax to use Checkpointer
-orbax==0.0.2
+orbax==0.0.2
+pytube