AI-103 Practice Question 99: SSML Prosody Emphasis
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 10 · question 9 of 10
A hospital's automated phone line reads medication instructions aloud using text-to-speech. Testers report that a caution ("take only one tablet, not two") is read in the same flat, uniform pace as the rest of the message, so callers sometimes miss it. The clinical wording itself must not change; only how that one phrase sounds should be different.
What should the team do?
Show answer and explanation
Answer:
SSML is the documented mechanism for exactly this control: it lets you "fine-tune your text to speech output attributes such as pitch, pronunciation, speaking rate, volume, and more," giving "more control and flexibility than plain text input." The <prosody> element's rate and pitch attributes slow down and change the pitch of a specific span of text, and <emphasis> marks a phrase for added stress, both scoped to the one caution, leaving the rest of the message and the underlying wording untouched.
- A is wrong because phrase lists bias *speech-to-text* recognition toward expected words; they have no effect on how text-to-speech paces or stresses synthesized output. - C is wrong because building and training a custom neural voice is a heavyweight, consent-gated undertaking to change a voice's identity; it doesn't control the delivery of one phrase in an existing voice, and it's disproportionate to the ask. - D is wrong because swapping voices changes the speaker's timbre and style for the entire call, not just the one caution, and doesn't guarantee that phrase specifically is read more slowly or with more stress than its surroundings.
Reference: Voice and sound with SSML