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.

• ✗

  Translating pre-recorded audio files overnight in a batch processing job

  Why it's wrong here

  This option describes an asynchronous, batch-oriented workload: translating pre-recorded audio files overnight is a file-processing job that runs on stored assets without user interaction. Real-time speech translation, by contrast, operates on a live audio stream, sending each utterance through speech recognition and translation with only a few hundred milliseconds of latency. The key differentiator is the temporal requirement — instant delivery of the translated output — which batch processing cannot fulfill. Azure offers batch transcription APIs for such offline scenarios, but they do not meet the definition of real-time translation.

• ✓

  Converting spoken words in one language to text or speech in another language instantly

  Why this is correct

  This is correct because real-time speech translation is specifically designed to process spoken input and produce translated text or synthesized speech with minimal delay. The pipeline works by streaming audio into a speech recognition engine, feeding the recognized text into a machine translation model, and optionally passing the result to a text-to-speech synthesizer for spoken output. The word "instantly" captures the low-latency, incremental nature of this process, which runs concurrently with the speaker's utterance rather than after the fact. Azure Speech Translation provides this capability through the Speech SDK for live conversations and presentations.

• ✗

  Generating subtitles for pre-existing videos stored in Azure Media Services

  Why it's wrong here

  This option describes an offline media-processing scenario, not real-time translation. Generating subtitles for pre-existing videos in Azure Media Services relies on batch video indexing and speech-to-text transcription of stored files, typically producing timed captions after the video has been fully processed. Real-time speech translation cannot operate on pre-recorded media because there is no live audio stream to consume incrementally; it works on speech as it is being produced. Even if subtitles were generated quickly, the underlying data source is a static file, which invalidates the core requirement of instantaneous live translation.

• ✗

  Converting text written in one language into spoken audio in the same language

  Why it's wrong here

  This option equates to a same-language text-to-speech operation, which does not involve any translation step. Converting written text into spoken audio in the same language is purely a synthesis task, where the input is text and the output is speech in that identical language. Real-time speech translation, however, requires two distinct languages: the source language of the spoken input and the target language of the output, whether text or speech. Furthermore, real-time translation typically starts from spoken words, not written text, so this option misses both the cross-language and speech-input aspects of the correct answer.

Go deeper

Related to this question