Create app.py

Created app.py

app.py

@@ -0,0 +1,66 @@
+import gradio as gr
+import whisper
+from pytube import YouTube
+
+def get_audio(url):
+    yt = YouTube(url)
+    return yt.streams.filter(only_audio=True)[0].download(filename="tmp.mp4")
+
+def get_transcript(url, model_size, lang, format):
+
+    model = whisper.load_model(model_size)
+
+    if lang == "None":
+        lang = None
+    
+    result = model.transcribe(get_audio(url), fp16=False, language=lang)
+
+    if format == "None":
+        return result["text"]
+    elif format == ".srt":
+        return format_to_srt(result["segments"])
+
+def format_to_srt(segments):
+    output = ""
+    for i, segment in enumerate(segments):
+        output += f"{i + 1}\n"
+        output += f"{format_timestamp(segment['start'])} --> {format_timestamp(segment['end'])}\n"
+        output += f"{segment['text']}\n\n"
+    return output
+
+def format_timestamp(t):
+    hh = t//3600
+    mm = (t - hh*3600)//60
+    ss = t - hh*3600 - mm*60
+    mi = (t - int(t))*1000
+    return f"{int(hh):02d}:{int(mm):02d}:{int(ss):02d},{int(mi):03d}"
+
+
+langs = ["None"] + sorted(list(whisper.tokenizer.LANGUAGES.values()))
+model_size = list(whisper._MODELS.keys())
+
+with gr.Blocks() as demo:
+
+    with gr.Row():
+
+        with gr.Column():
+
+            with gr.Row():
+                url = gr.Textbox(placeholder='Youtube video URL', label='URL')
+
+            with gr.Row():
+
+                model_size = gr.Dropdown(choices=model_size, value='tiny', label="Model")
+                lang = gr.Dropdown(choices=langs, value="None", label="Language (Optional)")
+                format = gr.Dropdown(choices=["None", ".srt"], value="None", label="Timestamps? (Optional)")
+
+            with gr.Row():
+                gr.Markdown("Larger models are more accurate, but slower. For 1min video, it'll take ~30s (tiny), ~1min (base), ~3min (small), ~5min (medium), etc.")
+                transcribe_btn = gr.Button('Transcribe')
+
+        with gr.Column():
+            outputs = gr.Textbox(placeholder='Transcription of the video', label='Transcription')
+
+    transcribe_btn.click(get_transcript, inputs=[url, model_size, lang, format], outputs=outputs)
+
+demo.launch(debug=True)