AI-900 Practice Question: Describe features of Natural Language Processing workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Speech to Text and Text Analytics API (standard)

  Why it's wrong here

  Although Speech to Text can handle the transcription portion, the standard Text Analytics API performs generic tasks like sentiment analysis and broad entity recognition (e.g., person, organization, location). It is not trained on clinical vocabularies and therefore cannot identify medical-specific entities such as drug dosages, ICD-10 codes, or symptom relationships. This option falls short because the analysis layer lacks the specialized medical models that Text Analytics for Health provides.

  When this WOULD be correct

  If the hospital only needed to transcribe conversations and perform general sentiment analysis or key phrase extraction (not medical-specific entities), then Speech to Text plus standard Text Analytics API would be sufficient.

• ✓

  Speech to Text and Text Analytics for Health

  Why this is correct

  Speech to Text provides high-quality real-time transcription of doctor-patient conversations, converting spoken language into text. The Text Analytics for Health service is then applied to that transcript, using healthcare-specific NLP models to extract entities such as diagnoses, medications, symptoms, and treatment plans. Together, they form a purpose-built pipeline for clinical documentation, which is exactly what the hospital needs.

• ✗

  Translator Text and Language Understanding (LUIS)

  Why it's wrong here

  Translator Text is designed to translate written text between languages, so it neither transcribes audio input nor extracts structured information from clinical speech. Language Understanding (LUIS) is a conversational tool that detects intents and custom entities from user utterances, but it requires extensive training on domain-specific phrases and is not inherently equipped for medical terminology. This pairing fails to deliver any speech-to-text capability and does not offer the out-of-the-box healthcare entity extraction needed by the hospital.

  When this WOULD be correct

  This option would be correct if the hospital needed to translate transcribed conversations from one language to another (e.g., Spanish to English) and then extract general intents (e.g., 'schedule appointment') from the translated text, without requiring medical-specific entities.

• ✗

  Speaker Recognition and Question Answering

  Why it's wrong here

  Speaker Recognition is a biometric service that identifies or verifies who is speaking based on voice characteristics, but it does not transcribe the content of the speech. Question Answering, such as QnA Maker, retrieves answers from a curated knowledge base by matching natural language queries, yet it has no capability to process raw audio or extract medical concepts from free-form text. Neither service addresses the requirement of turning spoken doctor notes into structured, medically relevant data.

  When this WOULD be correct

  A security system needs to verify the identity of doctors based on their voice during consultations, and then answer common medical questions from a FAQ database. Speaker Recognition for identity verification and Question Answering for FAQ retrieval would be appropriate.

Option-by-option analysis

Why each answer is right or wrong

Understanding why wrong answers are wrong — and when they would be correct — is what separates a 750 score from a 900. The AI-900 exam frequently reuses these exact scenarios with slightly different constraints.

Analysis generated from the official AI-900blueprint and verified against question context. The “when correct” sections are what AI assistants cite when candidates ask “what’s the difference between these options?”

Go deeper

Related to this question