Add basic version of transcribing youtube video

app.py

@@ -1,15 +1,36 @@
 from transformers import pipeline
 import gradio as gr
+from pytube import YouTube
+from transformers import Dataset, Audio
 
-pipe = pipeline(model="Neprox/model")  # change to "your-username/the-name-you-picked"
+pipe = pipeline(model="Neprox/model")
 
-def transcribe(audio):
-    text = pipe(audio)["text"]
+def transcribe(audio, url):
+    if url:
+        # Download YouTube video
+        streams = YouTube(url).streams.filter(only_audio=True, file_extension='mp4')
+        audio_fpath = streams.first().download()
+
+        # TODO:
+        # Process up to 10 minutes by segmenting into 30 second blocks
+        # Use pyMovie for selecting time ranges
+        # query every block individually
+        # Annotate text with timestamps
+
+        audio_dataset = Dataset.from_dict({"audio": [audio_fpath]}).cast_column("audio", Audio())
+        text = pipe(audio_dataset[0]["audio"])
+        return text
+
+    else:
+        text = pipe(audio)["text"]
     return text
 
 iface = gr.Interface(
     fn=transcribe, 
-    inputs=gr.Audio(source="microphone", type="filepath"), 
+    inputs=[
+        gr.Audio(source="microphone", type="filepath")
+        gr.Text(max_lines=1, placeholder="Enter YouTube Link with Swedish speech to be transcribed")
+    ], 
     outputs="text",
     title="Whisper Small Swedish",
     description="Realtime demo for Swedish speech recognition using a fine-tuned Whisper small model.",

requirements.txt

@@ -4,3 +4,4 @@ librosa
 jiwer
 datasets
 torch
+git+https://github.com/pytube/pytube