問題文
Design a voice-first assistant that answers from internal documents. Which arrangement covers the requirement end to end?
選択肢
- Send the audio directly to the index for retrieval so that no recognition step can introduce an error into the query that is used against the documents
- Generate the answer first from the most frequently asked questions and then recognize the user's speech to pick the closest one from that prepared set
- Recognize the speech and synthesize the transcript back to the user for confirmation, so the caller can correct a misheard question before any document is searched
- Recognize the speech, use the resulting text to retrieve from the index, generate the answer, and then synthesize that same answer back into spoken audio