AI Engineer: Designing and Implementing an Azure [AI-100] 4 — Questions and Answers
Question 1: You want to create a custom speech recognition model that handles domain-specific terminology accurately. What is the first step?
- Deploy a Standard Speech resource and increase the quota
- Provide custom training data such as transcripts and pronunciation files to Custom Speech (Correct answer)
- Switch to the S3 pricing tier of Speech Services
- Enable speaker diarization on the base model
Correct answer: Provide custom training data such as transcripts and pronunciation files to Custom Speech
Custom Speech lets you improve recognition accuracy by supplying domain-specific audio transcripts and pronunciation dictionaries as training data.
Question 2: You are using Azure Form Recognizer to extract data from invoices. Which prebuilt model should you select?
- The layout model
- The invoice prebuilt model (Correct answer)
- The receipt prebuilt model
- The business card prebuilt model
Correct answer: The invoice prebuilt model
Azure Form Recognizer provides a prebuilt invoice model trained to extract vendor, line items, totals, and other invoice-specific fields.
Question 3: Your solution uses Azure Bot Service connected to Microsoft Teams. Users report the bot does not respond in Teams. Where should you check first?
- The bot's App Service deployment logs
- Whether the Teams channel is enabled in the Bot Channels Registration (Correct answer)
- Azure Active Directory application permissions
- The bot's LUIS endpoint URL
Correct answer: Whether the Teams channel is enabled in the Bot Channels Registration
Each channel (Teams, Web Chat, etc.) must be explicitly enabled in the Bot Channels Registration for the bot to receive messages from it.
Question 4: You need to perform real-time translation of spoken audio from English to Spanish during a live call. Which combination of Azure services should you use?
- Speech-to-Text then Text Analytics then Text-to-Speech
- Speech Translation service with target language set to Spanish (Correct answer)
- Translator Text API with a Speech endpoint
- Custom Speech with a custom translation model
Correct answer: Speech Translation service with target language set to Spanish
Azure Speech Translation performs end-to-end spoken language translation in real time, handling speech recognition and translation in a single service.
Question 5: A Cognitive Search index includes a field populated by OCR during indexing. The extracted text quality is poor for scanned PDFs. What should you adjust?
- Increase the replica count of the search service
- Enable the imageAction property set to 'generateNormalizedImages' in the skillset (Correct answer)
- Add a synonym map to the index
- Switch the indexer to push mode
Correct answer: Enable the imageAction property set to 'generateNormalizedImages' in the skillset
Setting imageAction to 'generateNormalizedImages' improves OCR quality by normalizing image resolution before text extraction.
Question 6: You need to ensure your Azure Cognitive Services solution uses a customer-managed encryption key stored in Azure Key Vault. What must you configure?
- A private endpoint for the Cognitive Services resource
- Customer-managed keys (CMK) in the Cognitive Services resource encryption settings (Correct answer)
- A Managed Identity with Key Vault Secrets User role
- Data export to an encrypted Azure Storage account
Correct answer: Customer-managed keys (CMK) in the Cognitive Services resource encryption settings
Enabling CMK encryption in the Cognitive Services resource settings and linking it to an Azure Key Vault key fulfills the customer-managed encryption requirement.
Question 7: You trained an Azure Machine Learning pipeline and want to retrain the model automatically when new data arrives in Azure Blob Storage. What should you set up?
- A scheduled pipeline run using a cron expression
- An Azure Event Grid trigger that invokes the pipeline endpoint when a blob is created (Correct answer)
- An Azure Logic App that polls the storage account hourly
- A continuous deployment pipeline in Azure DevOps
Correct answer: An Azure Event Grid trigger that invokes the pipeline endpoint when a blob is created
Event Grid can detect blob creation events and trigger an Azure ML pipeline endpoint, enabling event-driven retraining without polling.
AI Engineer: Designing and Implementing an Azure [AI-100] 4
You want to create a custom speech recognition model that handles domain-specific terminology accurately.
What is the first step?