引導式測驗 · 中等 等級

語音系統中的語句結束偵測測驗

測試端點與語音活動、靜音閾值、暫停提示、過早中斷、回應延遲和恢復控制。

問題 1 的 8

What extra decision does endpointing make beyond detecting speech activity?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What extra decision does endpointing make beyond detecting speech activity?
    • 說話者的身份
    • Whether the speaker has finished the current utterance
    • The exact semantics of every word
    • The microphone’s hardware model
  2. Why can a short silence timeout feel fast but fail users?
    • It always preserves every final word
    • It removes room noise automatically
    • It establishes permission to act
    • It may cut off a mid-thought pause and lose later words
  3. What can pre-pause prosody add to a silence timer?
    • Evidence about whether a pause sounds final or continuing
    • A guaranteed transcript of unspoken words
    • A direct measurement of user intent
    • A separate microphone channel
  4. Which pair of measures captures the core endpointing tradeoff?
    • Camera accuracy and image size
    • Only transcript word count
    • Premature-cutoff rate and post-speech response latency
    • Only assistant output volume
  5. A noise source keeps a voice system listening after the user stops. What should be tested?
    • Only the app icon appearance
    • Speech-versus-background discrimination in that environment
    • Only labels in an image dataset
    • Whether every pause means consent
  6. Why might a dictation app choose a longer endpoint wait than a simple command interface?
    • A writer may pause while continuing a multi-sentence thought
    • Longer waits automatically improve word accuracy
    • Dictation contains no background noise
    • Simple commands never include a pause
  7. Which design helps a person recover after a premature cutoff?
    • Treat the cutoff as irreversible
    • Delete every partial transcript
    • Execute every provisional command
    • Let them continue or correct the interrupted request
  8. What does silence alone not prove in a consequential voice flow?
    • That acoustic energy was low for a period
    • That a silence threshold was reached
    • That the user consented or intended a specific action
    • That a microphone remained active