69 lines (57 loc) · 3.08 KB

Previous | Table of Contents | Next

16. Quick Reference Tables

API Key Quick Reference

Header Name	Use
`Ocp-Apim-Subscription-Key`	Cognitive Services resource key (Key1/Key2) — used across many REST endpoints
`api-key`	OpenAI endpoints — use your Azure OpenAI resource key
`Prediction-Key`	Custom Vision prediction key (portal/endpoint)

File Size / Limits Summary

Service	Max Size / Notes
Document Intelligence	50 MB per file (500 MB via storage pipelines)
Custom Vision	~6 MB single upload; 64 images per batch guidance
Language Services	~5,120 chars per doc; 1,000 docs per batch; ~1 MB batch
Speech (batch)	Guidance ~200 MB
AI Search	16 MB per document (JSON)
Video Indexer	Limits vary; for large videos use Blob + `videoUrl` (see docs)

Video File Outputs (Video Indexer)

File / Artifact	What it’s for
`insights.json`	Full analysis results: transcript segments, speakers, OCR, faces, topics/keywords, timestamps, confidence
`transcript.vtt` / `transcript.srt`	Native-language captions for the detected/specified spoken language
`translated_<lang>.vtt` / `translated_<lang>.srt`	Translated captions for requested target language(s)
Thumbnails / keyframes (`.jpg`)	Visual navigation, previews; also referenced by insights metadata

Model Selection (cheat sheet)

Task	Model
Images	DALL‑E
Speech → text	Whisper / Azure Speech STT
Chat	GPT‑3.5‑Turbo / GPT‑4
Embeddings	text-embedding-ada-002 (or current recommended model)

AnalyzeTextAsync() Snapshot

Aspect	Details
Purpose	sentiment, entities, key phrases, PII, language detection
Key options	`includeOpinionMining`, `modelVersion`, `showStats`, `disableServiceLogs`
Returns	`sentiment`, `confidenceScores`, `sentences[]`, `warnings[]`, `statistics`

Token Behavior Summary

Scenario	finish_reason	Behavior
Completes	`stop`	Full response returned
Hits max_tokens	`length`	Response truncated
Content policy hit	`content_filter`	Blocked/filtered

Confidence Score Thresholds (guidance)

Service	Low	Medium	High
Document Intelligence	< 0.7	0.7–0.89	≥ 0.9
Language Detection	< 0.7	0.7–0.84	≥ 0.85
Custom Vision	< 0.6	0.6–0.79	≥ 0.8

JSONPath Wildcard Examples

Pattern	Finds
`$..confidence`	All confidence values anywhere in JSON
`$.fields.*.confidence`	Top-level extracted field confidences
`$.predictions[*].probability`	Prediction probabilities (Custom Vision)

Previous | Table of Contents | Next