A. Correct.
Azure AI Content Safety and Azure AI Vision can analyze images for unsafe or inappropriate content.

B. Incorrect.
DNS and Traffic Manager route network traffic and do not moderate images.

C. Incorrect.
Backup and Site Recovery protect workloads and do not analyze content.

D. Incorrect.
Files and queues store data and messages but do not provide image moderation intelligence.

A. Incorrect.
A NAT gateway controls outbound network connectivity and does not expand search terms.

B. Incorrect.
Blob lifecycle rules manage storage tiers and deletion.

C. Correct.
Synonym maps expand queries to equivalent terms in Azure AI Search.

D. Incorrect.
Availability sets improve VM placement availability and are unrelated to search.

A. Incorrect.
DNS settings do not configure speech translation.

B. Correct.
Speech translation needs the source recognition locale and one or more target translation languages.

C. Incorrect.
Sentiment analysis does not translate speech.

D. Incorrect.
A SAS token grants storage access and does not configure Speech SDK languages.

A. Incorrect.
Custom Vision detects image objects and does not tune video transcript vocabulary.

B. Incorrect.
Firewall DNAT is networking and does not affect transcript recognition.

C. Incorrect.
Blob versioning preserves objects and does not customize speech transcription.

D. Correct.
Language model customization can improve transcript recognition for specific terms by adding them to the include list.

A. Incorrect.
show-all-intents returns additional prediction output but does not log utterances for learning.

B. Correct.
Logging user queries enables review of endpoint utterances and supports active learning.

C. Incorrect.
dns=true is not a language model learning setting.

D. Incorrect.
Cache headers do not capture training utterances.

A. Incorrect.
A DNS resolver handles name resolution and not semantic similarity.

B. Incorrect.
Speech synthesis produces audio and does not create text embeddings.

C. Correct.
Embedding models produce vectors that can be compared with cosine similarity.

D. Incorrect.
WAF rules protect web apps and do not generate vectors.

A. Incorrect.
Traffic Manager routes DNS traffic and does not process documents.

B. Incorrect.
Firewall Policy secures network traffic and does not extract document tables.

C. Incorrect.
Disk Encryption protects VM disks and is unrelated to document extraction.

D. Correct.
Azure AI Document Intelligence extracts text, tables, key-value pairs, and fields from documents.

A. Correct.
Omnichannel for Customer Service can manage live agent engagement and handoff experiences.

B. Incorrect.
Blob archive tier stores infrequently accessed data and does not manage agents.

C. Incorrect.
Private DNS resolves names and does not provide customer service handoff.

D. Incorrect.
Load Testing evaluates performance and does not manage support agents.

A. Incorrect.
Publishing every output ignores the confidence threshold requirement.

B. Correct.
A human-in-the-loop review process lets people correct or approve low-confidence AI output.

C. Incorrect.
Disabling confidence scores removes the signal needed for routing to review.

D. Incorrect.
Anonymous access weakens security and does not improve AI responses.

A. Incorrect.
Owner for every user is excessive and violates least privilege.

B. Incorrect.
Shared keys in documentation are insecure and hard to govern.

C. Correct.
RBAC assignments to groups at the smallest practical scope support least privilege and manageable access.

D. Incorrect.
Anonymous access does not meet secured RBAC requirements.