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.

• ✗

  Checking whether a text-to-speech voice pronounces technical terms correctly

  Why it's wrong here

  Verifying whether a text-to-speech voice pronounces technical terms correctly involves managing pronunciation lexicons or SSML phoneme settings, which is a synthesis configuration task, not a speech evaluation task. Pronunciation assessment requires a human speaker's audio stream and compares it to a reference transcript to produce accuracy, fluency, and completeness scores. TTS output is machine-generated speech, so it lies outside the scope of a feature designed to score human pronunciation.

• ✓

  Scoring a speaker's accuracy, fluency, and completeness against native pronunciation norms

  Why this is correct

  This describes the core function of Azure's pronunciation assessment: it ingests spoken audio of a human speaker, aligns it to a phonetic transcription of the expected sentence, and returns overall and per-phoneme scores for accuracy (phonetic match), fluency (rate and pauses), and completeness (fully uttered words). These scores reflect how closely the speaker's production approximates a native reference, which is why the feature supports language learning and speech therapy scenarios. This is precisely the definition, so it is correct.

• ✗

  Detecting regional accents to route calls to the appropriate customer service team

  Why it's wrong here

  Accent-based call routing identifies a caller's dialect or regional accent and uses that classification to select a suitable customer service queue, which is a recognition-driven classification workflow. In contrast, pronunciation assessment quantifies how closely a speaker's articulation matches standard native norms by scoring phoneme-level accuracy, speaking rate, and word completeness. Accent detection focuses on categorizing speech variation, whereas pronunciation assessment focuses on measuring distance from an ideal target, so the described routing scenario does not apply.

• ✗

  Generating a list of commonly mispronounced words in a specific language

  Why it's wrong here

  Generating a list of commonly mispronounced words in a specific language is a static linguistic analysis that aggregates lexicon data, not an evaluation of a particular speaker's audio. Azure's pronunciation assessment runs in real time on an utterance, scoring individual phonemes against a native reference, and does not produce cross-speaker lists of error-prone vocabulary. Therefore it is a reference resource, not a scoring function for pronunciation proficiency.

Go deeper

Related to this question