+ choosing model size

app.py

@@ -11,9 +11,11 @@ AUTO_DETECT_LANG = "Auto Detect"
 def transcribe(audio, state={}, model_size='base', delay=1.2, lang=None, translate=False):
     time.sleep(delay - 1)
 
+    global current_size
+    global model
     if model_size != current_size:
-        model = whisper.load_model(model_size)
         current_size = model_size
+        model = whisper.load_model(current_size)
   
     transcription = model.transcribe(
         audio,
@@ -35,11 +37,11 @@ def transcribe(audio, state={}, model_size='base', delay=1.2, lang=None, transla
   
 
 title = "OpenAI's Whisper Real-time Demo"
-description = "A simple demo of OpenAI's [**Whisper**](https://github.com/openai/whisper) speech recognition model."
+description = "A simple demo of OpenAI's [**Whisper**](https://github.com/openai/whisper) speech recognition model. This demo runs on a CPU. For faster inference choose 'tiny' model size and set the language explicitly."
 
 model_size = gr.Dropdown(label="Model size", choices=['base', 'tiny', 'small', 'medium', 'large'], value='base')
 
-delay_slider = gr.inputs.Slider(minimum=1, maximum=5, default=1.2, label="Rate of transcription, s")
+delay_slider = gr.inputs.Slider(minimum=1, maximum=5, default=1.2, label="Rate of transcription")
 
 available_languages = sorted(tokenizer.TO_LANGUAGE_CODE.keys())
 available_languages = [lang.capitalize() for lang in available_languages]