AI Bot
AI-200: Vision & Speech Services Flashcards
Microsoft Azure AI Cloud Developer Associate AI-200 Flashcards
Study our AI-200: Vision & Speech Services flashcards for the Microsoft Azure AI Cloud Developer Associate AI-200 exam with 47+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| How do you authenticate to Cognitive Services using Azure AD | Request an AAD token and use it in Authorization header Bearer token |
| How do you call Custom Vision prediction at scale | Batch prediction endpoint or export model for edge inference |
| How do you evaluate OCR accuracy | Use metrics like word error rate WER character error rate CER and field level accuracy |
| How do you export Custom Vision models | Export to platforms like ONNX TensorFlow CoreML or as Docker for edge deployment |
| What are core capabilities of Azure Computer Vision | Image analysis OCR object detection image tagging and spatial analysis |
| What does Face API offer | Face detection attributes recognition verification and identification |
| What header sets content type for image upload | Content Type like application octet stream or image jpeg |
| What image size considerations improve model performance | Resize keeping aspect ratio avoid extreme upscaling and ensure objects are large enough |
| What is a confidence score in speech to text | Probability estimate the model assigns indicating transcription reliability |
| What is a pronunciation lexicon | Custom list of words and phonetic pronunciations to improve recognition of rare or domain specific terms |
| What is a subscription key | Static API key used to authenticate requests to legacy cognitive services endpoints |
| What is anchor box concept in object detection | Predefined bounding box shapes used to detect objects at different scales and aspect ratios |
| What is automatic punctuation | Post processing to insert punctuation into transcribed text |
| What is content moderation in Vision | Detecting adult or racy content and unwanted imagery for safety and compliance |
| What is Custom Vision main purpose | Train custom image classifiers and object detectors on labeled images |
| What is end to end latency for real time speech factors | Network latency audio capture encoding model inference and response delivery |
| What is endpoint regionality importance | Use resources in the same region to reduce latency and comply with data residency |
| What is face identification | Find a person in a database using a detected face and person group or large person group |
| What is face verification | Compare two faces to determine if they belong to the same person |
| What is Form Recognizer custom model | Train models to extract fields from custom document layouts with labeled samples |
| What is Form Recognizer layout model | Extract low level elements like text lines words tables and selection marks |
| What is Form Recognizer prebuilt model used for | Extract common document types like receipts invoices and identity documents |
| What is model iteration in Custom Vision | Each training run produces an iteration you can evaluate compare and promote to production |
| What is neural voice in TTS | Advanced voice model with more natural prosody and expressiveness than standard voices |
| What is rate limiting and how to handle it | Service limits on requests per second use retries exponential backoff and batching |
| What is speaker adaptation in Custom Speech | Customize acoustic and language models to improve recognition for specific voices or domain vocabulary |
| What is speaker diarization | Labeling segments of audio with speaker identities for multi speaker transcripts |
| What is speech translation | Real time or batch conversion of spoken language from one language to another with text or audio output |
| What is SSML used for in TTS | Control prosody pronunciation pauses and voice selection in speech synthesis |
| What is table extraction challenge | Detecting rows columns spanning cells and irregular table structures |
| What is Text to Speech main function | Convert text or SSML to spoken audio using neural or standard voices |
| What is the benefit of endpoint vs SDK | Endpoint is raw REST access SDK provides higher level helpers with authentication retries and streaming support |
| What is the common pattern for long running operations | Initial POST request returns operation location then poll GET until status is succeeded |
| What is the difference between classification and object detection | Classification assigns labels to whole images object detection finds and localizes objects with bounding boxes |
| What is the difference between person group and large person group | Person group is for smaller sets large person group supports scalable indexing and retrieval |
| What is the recommended sample rate for speech recognition | 16000 Hz mono PCM for best accuracy on general speech |
| What is transfer learning in Custom Vision | Starting from pretrained models and fine tuning on your labeled dataset |
| What is voice activity detection VAD | Automatic detection of speech segments to remove silence and reduce processing |
| What preprocessing helps OCR accuracy | Deskewing binarization contrast enhancement and cropping to text regions |
| What preprocessing is used for face recognition | Align faces crop to landmarks and normalize scale and illumination |
| When should you use Speech to Text batch | For long recordings or offline transcription using batch transcriptions |
| When should you use Speech to Text real time | For live streaming audio like calls or interactive apps using WebSocket or SDK streaming |
| Why are bounding boxes important in Vision | To localize objects allow spatial reasoning and enable cropping for downstream tasks |
| Why choose real time processing | When low latency responses are required for interactive applications |
| Why convert audio to mono before recognition | Most models expect single channel audio and mixing prevents channel mismatch |
| Why perform noise reduction before speech recognition | To improve model accuracy by removing background noise and artifacts |
| Why use batch processing for vision tasks | To process many items efficiently with asynchronous jobs and lower per item overhead |
About the Flashcards
These Flashcards for the Microsoft Azure AI Cloud Developer Associate exam provide an excellent resource for students reviewing essential cloud AI services and architecture concepts. The deck focuses heavily on helping you master key terminology and technical ideas related to cognitive services. Students can study fundamental concepts such as computer vision capabilities, custom image classification, and object detection workflows.
Additionally, the collection covers vital speech processing concepts, including text-to-speech synthesis, acoustic preprocessing, and transcription models. You will also find useful review material on API authentication, long-running operation patterns, and evaluating machine learning accuracy. By using this deck, you can effectively reinforce your knowledge of document intelligence and facial recognition systems before test day.
Topics covered in this flashcard deck:
- Azure Computer Vision and OCR
- Custom Vision and object detection
- Form Recognizer and document layout
- Face API and recognition capabilities
- Speech processing and transcription
- API authentication and performance limits