6.2. Reflection Checkpoint

💡 First Principle: Speech and translation are modality and language bridges around the text core. Speech-to-text and text-to-speech wrap an agent's text reasoning so it can listen and talk; translation moves text or speech across languages. The exam names both as distinct objectives under text analysis — they are not folded into generic "language tasks."

Why care: the official outline lists speech-to-text and text-to-speech for agentic interactions, speech as an agent modality including custom speech models, multimodal reasoning from audio inputs, and speech translation — plus text translation via Azure Translator in Foundry Tools or LLM-powered flows. A study plan built only on classification and sentiment misses these named sub-skills.

⚠️ Common Misconception: "A voice agent needs a special voice model that hears and talks." The standard pattern keeps a text reasoning core and brackets it: speech-to-text on the way in, text-to-speech on the way out. The reasoning, grounding, and evaluation all happen in text, which is what keeps a voice agent observable and groundable.