引導式測驗 · 硬 等級
呼叫中心測驗中的人工智慧口音中和和語音翻譯
區分語音辨識、翻譯和口音轉換,同時測試同意、評估和隱私保護。
問題 1 的 8
A system converts a caller’s speech into text but does not render another language. Which capability is it performing?
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。
- A system converts a caller’s speech into text but does not render another language. Which capability is it performing?
- Speech-to-speech translation
- Speech recognition or transcription
- Accent conversion only
- Speaker consent verification
- Why should a call center distinguish translation from accent conversion?
- They solve different tasks and have different error and consent implications
- They are guaranteed to be the same technology
- Accent conversion always translates between languages
- Translation never changes audio
- A pilot asks agents to use voice conversion, but declining may reduce their shifts. What is the central problem?
- The system has too many supported languages
- The call recordings are too short
- The customer can hear translated captions
- Participation is not meaningfully voluntary
- A translated call sounds natural, but the customer disputes a changed order number. What evaluation would catch this risk?
- Measure only how natural the synthesized voice sounds
- Count how many calls the software receives
- Have qualified listeners assess semantic accuracy, including names, negation and numbers
- Compare the agent’s accent before and after conversion
- Agents and callers struggle to understand one another across languages. Which option can support comprehension without changing an agent’s voice?
- Require every agent to change their accent
- Offer translated captions or interpreter access for the languages involved
- Hide language preferences from the caller
- Use a voice clone without notifying anyone
- Why can speech-to-speech translation errors compound?
- Recognition, translation and synthesis can each introduce a different error
- The stages automatically correct every earlier mistake
- Only the final audio is processed
- A fluent voice guarantees exact wording
- Which use should be avoided when transformed audio may contain errors?
- Offering it as an optional accessibility feature
- Testing performance under noisy call conditions
- Providing a caption alongside synthesized speech
- Using it as the sole basis for discipline or quality scoring
- What information should a transparent voice-processing notice explain?
- Only the product’s brand name
- That the model is always correct
- What is transformed, whether audio is stored, who can access it and available alternatives
- The agent’s private performance rating