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.

• ✗

  Creating a custom voice persona that sounds different from the standard Azure voices

  Why it's wrong here

  Creating a custom voice persona that sounds distinct from standard Azure voices describes Custom Neural Voice, which is a text-to-speech feature that synthesizes unique artificial voices for playback. Custom Speech, by contrast, focuses exclusively on speech recognition (speech-to-text), converting spoken audio into text, not generating spoken output. The 'custom' in Custom Speech refers to customizing the recognition model's understanding of your audio, not customizing the speaking voice. Thus, this option mistakes a speech synthesis capability for a speech recognition capability.

• ✓

  Fine-tuning speech recognition for domain-specific vocabulary, accents, or noisy environments

  Why this is correct

  Custom Speech in Azure AI Speech is precisely the feature for improving speech-to-text accuracy on domain-specific jargon, unique accents, or noisy audio environments. It works by training custom acoustic models on labeled audio and custom language models on domain-specific text, thereby adapting the recognizer to vocabulary and speaking styles that generic models miss. In Azure this is also exposed through the 'Custom Speech' portal and Speech Studio, and it requires your own training data. This is the correct definition of Custom Speech's core value.

• ✗

  Configuring speech recognition to only accept voice commands from authorised users

  Why it's wrong here

  Speaker verification, not speech recognition adaptation, is the Azure feature that restricts voice commands to authorised users. This scenario describes speaker identification/verification, which compares a speaker's unique vocal characteristics against enrolled profiles. Custom speech, in contrast, fine-tunes the acoustic and language models to understand the content of what is said, not to verify who is saying it. Therefore, this option conflates the distinct Azure AI Speech services of speaker recognition and custom speech.

• ✗

  Building a custom programming language for writing speech processing scripts

  Why it's wrong here

  Writing a domain-specific language for speech processing is a software engineering task that does not reflect Azure Custom Speech's purpose or capabilities. Custom Speech is a managed model-customization service that lets you upload audio and transcription data to adapt Microsoft's pre-trained recognition model. It does not involve authoring programming languages or writing scripts that process audio programmatically. This choice confuses the act of creating custom code with the act of training a custom speech recognition model.

Go deeper

Related to this question