Implement computer vision solutions 3.2 Design and implement multimodal understanding workflows
A news publisher wants each broadcast recording split into chapters by story, with a…
Study guide skill: Implement video analysis workflows to process and interpret video segments
Question B1-035
A news publisher wants each broadcast recording split into chapters by story, with a transcript and a short description for each part. Which solution fits with the least custom code?
Select one answer.
Answer: C
Content Understanding for video returns inline transcripts, natural-language segment descriptions and automatic segmentation, and you can define custom segments, for example to split a news broadcast into chapters by story. Batch transcription only produces text, Sora generates video, and semantic ranker reranks search results.
Deep dive
How it works
Azure Content Understanding for video produces a standard set of video metadata and lets you add custom fields and segments using generative models. The prebuilt video analyzer returns RAG-ready output: inline transcripts in WEBVTT format, ordered key frames, natural-language segment descriptions that combine visual and speech context, and automatic segmentation. The output can go straight into a vector store with no post-processing.
Customization is what fits this scenario. Microsoft's own example of custom segments is splitting a news broadcast into chapters by the topics or stories discussed. Custom fields can capture things like the products and brands that are seen or mentioned. Internally the service works in two stages. Content extraction captures foundational metadata such as transcripts and shots. Field extraction then uses a generative model to produce custom fields and perform segmentation. Broadcast media is one of the listed use cases.
Note the label: the video overview marks its current API version as public preview, without a service-level agreement and not recommended for production workloads.
The other options, and when you'd use them
- Azure AI Search semantic ranker: Semantic ranker reranks search results using Microsoft's language understanding models and can return captions and answers. Use it to improve relevance once content is already indexed. It doesn't analyze video.
- Sora 2 video remix: Sora 2 (preview) creates video from text, images or video. Remix makes targeted changes to an existing generated video instead of regenerating it. Use it for creating video, not analyzing recordings. Video generation also uses a preview API version.
- Azure AI Speech batch transcription only: Batch transcription transcribes a large amount of stored audio asynchronously. You'd get transcripts, but no story chapters or descriptions without building that logic yourself.
Exam tip
"Split video into scenes or chapters, with transcript and descriptions, and little code" means a Content Understanding video analyzer with custom segments. "Make a video" means Sora, and "rank search results" means semantic ranker.
References
Microsoft Learn references
Last reviewed: