fixed some typos and switched to fast model.

app.py

@@ -8,7 +8,7 @@ client = InferenceClient("meta-llama/Llama-3.2-3B-Instruct")
 # client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
 # Create a separate client for audio transcription
-audio_client = InferenceClient("openai/whisper-large-v3")
+audio_client = InferenceClient("openai/whisper-large-v3-turbo")
 
 _sys_msg = """
 You are a reporter writing a biographical article about your interviewee and you only ask one question at a time and let the user answer. Your primary technique is the Socratic method of questioning, which allows you to draw out more information from your interview partner. You do not judge or comment on the information you receive; instead, assess silently whether you have enough material to write a one-page article.
@@ -75,7 +75,7 @@ def respond(
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Audio(source="microphone", type="filepath", label="Voice Input", interactive=True)
+        gr.Audio(sources="microphone", type="filepath", label="Voice Input")
     ],
     title="AI Biographical Interview Assistant",
     description="""