Implement information extraction solutions 5.1 Build retrieval and grounding pipelines

You index recorded webinars, each in English or Spanish, with a Content Understanding…

Study guide skill: Ingest and index content, such as documents, images, audio, and video

Question B5-023

You index recorded webinars, each in English or Spanish, with a Content Understanding audio analyzer before loading the output into a RAG index. Which two configuration choices follow the product guidance? (Choose two.)

Select 2 answers.

A Define a FullTranscript field with method extract
B Don't define a field just to hold the transcript, because the transcript is already in the analyzer output ✓ Correct answer
C Set enableSegment to true on the audio analyzer to split each webinar into chapters
D Leave locales empty so the service considers all languages, which gives the best accuracy
E Set locales to only the English and Spanish locales you expect ✓ Correct answer

Answer: B, E

Specify only the languages you expect, because including too many languages increases transcription errors, and specifying multiple locales turns on language detection. Speech transcripts are automatically available in analyzer output, so define fields only when you need extra processing such as summarization. Audio analyzers support only the returnDetails and locales configuration options, and the extract method is only supported for document analyzers.

Deep dive

How it works

Two Content Understanding best practices decide this audio setup.

1. Narrow the languages. Specify only the languages you expect. Including too many increases transcription errors, because the system has to guess which language is being spoken. The best-practices page uses this exact case: for English and Spanish content, configure only those two rather than autodetecting from all languages. In the audio analyzer you set this with locales, for example en-US and es-ES. The audio overview's locale table shows that with multiple locales you get single-language transcription based on language detection, which fits one-language-per-webinar files.

2. Don't re-extract the transcript. Speech transcripts are automatically available in analyzer output. Define fields only when you need extra processing, such as a summary or entities. The transcript lands in the markdown output you load into the RAG index anyway.

Audio analyzers also have a short option list: the analyzer reference shows only returnDetails and locales for prebuilt-audio. Plan around that.

The other options, and when you'd use them

A FullTranscript field with method extract: Duplicates content you already get. And extract is only supported for document analyzers.
enableSegment on the audio analyzer: Audio analyzers support only the returnDetails and locales options; enableSegment is listed for video analyzers.
Leave locales empty for best accuracy: The guidance says the opposite. Too many candidate languages increases transcription errors.

Exam tip

Content Understanding audio: set locales to just the expected languages, and don't define fields that only copy the transcript. Audio options = returnDetails + locales.

References

Last reviewed: 10 Oct 2026

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