引導式測驗 · 中等 等級

AI 口述歷史逐字稿轉錄測驗

測試人工智慧口述歷史轉錄:草稿與來源音訊、分類、方言、敘述者同意、轉錄審查、保存和訪問

問題 1 的 8

當 AI 轉錄與錄音衝突時,哪個來源應作為修正指引?

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. 當 AI 轉錄與錄音衝突時,哪個來源應作為修正指引?
    • The model’s most fluent sentence
    • A summary generated from the transcript
    • The original audio recording
    • A guess based on the interview topic
  2. 為何應在可能時提供旁白者轉錄審閱?
    • It guarantees the speech model is perfect
    • It replaces the need to preserve audio
    • It permits the editor to rewrite the interview freely
    • They can correct names, context, privacy choices, and how their words are represented
  3. 當音訊不清時,語言模型上下文能做什麼?
    • Produce a plausible completion that was not actually spoken
    • Guarantee the exact words
    • Remove all speaker overlap
    • Preserve every pause
  4. 編輯應如何處理聽不清的段落?
    • Insert the most likely phrase without notation
    • Mark uncertainty rather than silently guessing from context
    • Delete the audio segment
    • Ask the model to complete it confidently
  5. 在將受限訪談送至雲端轉錄服務前,應檢查哪些項目?
    • Whether the tool can export time-stamped text, without checking how it handles uploaded audio
    • Whether the service appears on an approved list, without reviewing its retention or training terms
    • Consent terms and vendor retention, access, training, and deletion practices
    • Whether the vendor uses a modern website design
  6. Why preserve the original recording alongside corrected transcripts?
    • The transcript is always more authoritative
    • The audio can be discarded after transcription
    • The recording proves every memory is historically accurate
    • The recording lets future users verify wording and interpretation
  7. Speaker diarization can fail in which common way?
    • Overlapping voices or short turns can be assigned to the wrong speaker
    • Diarization changes the recording’s sample rate
    • It proves the identity of each voice
    • It always preserves pauses accurately
  8. What does OHA’s guidance emphasize about interview use and release?
    • Automatic public release after transcription
    • Informed consent, transparency, and clear parameters for preservation and access
    • A ban on all technology use
    • A transcript that omits narrator preferences