Kuis terkait panduan · Sedang Tingkat

Kuis Agen Suara Waktu Nyata

Uji kehati-hatian transkrip parsial, latensi end-to-end, pembatalan barge-in versus pembatalan alat, rekonsiliasi tindakan dengan batas waktu habis, handoff, dan privasi audio.

Jalur panduan terkaitAgen Suara Waktu Nyata
Pertanyaan 1 dari 8

A voice agent hears only a provisional partial phrase. Which action should wait?

Pratinjau setiap pertanyaan8 pertanyaan

Jawaban dan penjelasan muncul saat Anda mengikuti kuis.

  1. A voice agent hears only a provisional partial phrase. Which action should wait?
    • Displaying provisional captions
    • A consequential account or record change until interpretation is stable and confirmed
    • Showing a listening indicator
    • Measuring response latency
  2. Why measure more than the language model’s response time?
    • Endpointing, recognition, retrieval, tools and playback all add delay
    • Model time always equals user-perceived delay
    • Audio playback occurs before user speech
    • Only server clock speed matters
  3. A user says “Stop” while an appointment update is underway. What cannot be inferred from silent playback?
    • That audio output stopped
    • That the user interrupted
    • That the agent should check action state
    • That the external update was cancelled
  4. A tool call times out after a spoken request. What should happen before the agent retries it?
    • Repeat immediately because timeouts always roll back
    • Tell the user it certainly failed
    • Reconcile whether the external operation already completed
    • Delete the action log
  5. Which state should a voice agent track for a write operation?
    • Only whether the voice sounded friendly
    • Pending, completed or impossible to reverse
    • Only the model’s token count
    • Only the microphone volume
  6. A task cannot be completed after connection loss. What response fits the guide?
    • Claim success to preserve conversational flow
    • Hide the failure and retry indefinitely
    • Assume a transcript means the tool succeeded
    • Honest status and a usable handoff or recovery path
  7. Why test overlapping speakers and ambiguous names?
    • They are impossible in real conversations
    • A fast response guarantees they are resolved
    • They can change the interpretation on which the agent acts
    • Only transcription display is affected
  8. Which data deserve access and retention controls?
    • Recordings, transcripts and action logs under the actual policy
    • Only the final audio waveform
    • Only the UI theme
    • Only the user’s chosen voice name