Implement text analysis solutions 4.2 Implement speech solutions

A language-learning app asks students to read a passage aloud for a few minutes and must…

Study guide skill: Enable multimodal reasoning from audio inputs

Question B7-018

A language-learning app asks students to read a passage aloud for a few minutes and must flag words they skipped or added, using Speech pronunciation assessment with a reference text. How should the app get omissions and insertions?

Select one answer.

  • A Use continuous mode and compare the recognized words with the reference text to find omissions and insertions ✓ Correct answer
  • B Use continuous mode with EnableMiscue set to true so the service returns omission and insertion tags
  • C Use continuous mode; omission and insertion tags are returned automatically without extra work
  • D Use single-shot recognition with EnableMiscue set to true on the whole recording

Answer: A

For audio longer than 30 seconds, Microsoft says to use continuous mode. In continuous mode, the EnableMiscue option isn't supported, so to get omission and insertion tags you compare the recognized results with the reference text yourself.

Deep dive

How it works

Pronunciation assessment scores how a speaker reads a reference text. Miscues are words the speaker left out (Omission) or added (Insertion), and the per-word ErrorType can report them.

Normally you get those tags with the EnableMiscue setting: it enables miscue calculation when the pronounced words are compared to the reference text, and when it's True, ErrorType can be set to Omission or Insertion.

Two rules change the picture for a few-minute reading:

  1. If your audio exceeds 30 seconds, use continuous mode.
  2. In continuous mode, the EnableMiscue option isn't supported. To get Omission and Insertion tags, you compare the recognized results with the reference text yourself.

So the app runs continuous recognition, collects the recognized words, and aligns them with the reference text to find skipped and added words. Microsoft publishes continuous-mode samples on GitHub.

The other options, and when you'd use them

  • Continuous mode with EnableMiscue true: EnableMiscue isn't supported in continuous mode, so the service won't return the tags.
  • Tags returned automatically: In continuous mode you must compare the recognized results with the reference text to get them.
  • Single-shot with EnableMiscue: Audio over 30 seconds should use continuous mode, and these recordings run for minutes.

Exam tip

Short clip = EnableMiscue works. Over 30 seconds = continuous mode, no EnableMiscue, compare with the reference text yourself.

References

Last reviewed:

Ready to test yourself? The quiz asks this and every other AI-103 question, with instant feedback. Practice this in a quiz →

All Text analysis questions · AI-103 study guide

Thanks for your feedback!