Last Attempt AI-102 Questions

To detect harmful content with Azure AI Content Safety using the .NET SDK, you instantiate a ContentSafetyClient with the endpoint and key. For text moderation, you construct an AnalyzeTextOptions object with the input string (and optionally specify categories, blocklists, etc.). You then call client.AnalyzeText(AnalyzeTextOptions options) and inspect the result (e.g., categories analysis and blocklist matches).

Code completed:

var client = new ContentSafetyClient(new Uri(endpoint), new AzureKeyCredential(key));

var request = new AnalyzeTextOptions( " what is the weather forecast for Seattle " );

Response < AnalyzeTextResult > response = client.AnalyzeText(request);

Microsoft Azure AI References

Azure AI Content Safety – SDK usage for text analysis with ContentSafetyClient and AnalyzeTextOptions (Analyze Text API).

Azure AI Content Safety – Concepts: text categories (hate/sexual/violence/self-harm) and blocklists, returned in AnalyzeTextResult .

Comprehensive Detailed Explanation

A custom neural voice (professional voice) project in Azure AI Speech requires that you follow Responsible AI and consent steps before any model training. The correct workflow in Speech Studio / Azure AI Foundry is:

Create a custom voice project – This is the container that will hold the voice talent consent, your training datasets, models, and deployments. Each project is scoped to a locale and other attributes. Microsoft Learn

Collect and upload consent – Microsoft requires explicit authorization from the voice talent. In practice, you add the voice talent consent to the project. The platform requires the verbal consent recording (WAV); many guidance and exam items also include the signed consent document (PDF) as part of consent capture to meet Responsible AI expectations. The consent step is performed before uploading training data.

Upload the voice talent’s verbal consent as a WAV file – The consent audio verifies that the person who granted consent is the same as the speaker in your training data. The documentation shows consent is a recorded statement uploaded to the project. Microsoft Learn

Analyze the quality of the audio data and fix issues – After data upload, Speech Studio runs validation and provides a per-utterance analysis (format checks, sampling rate, script match, pronunciation/noise scoring). You must resolve issues and run Analyze data before the dataset can be used for training. Microsoft Learn

Train the model using a neural training method – Start model training from the project and choose Neural as the training type for custom neural voice. Microsoft Learn

Why not “upload speech samples as WMA/MP3 files”?

For the standard, recommended Individual utterances + matching transcript data type used to fine-tune a professional neural voice, the audio files must be WAV/RIFF PCM; the docs explicitly reference “.wav” throughout data checks and corrections. WMA isn’t supported, and MP3 isn’t accepted for this core data type. (Long audio / audio-only workflows can accept MP3 for preprocessing, but the canonical training dataset for custom neural voice is WAV.)

Microsoft Azure AI References

Create a project for professional voice (Speech Studio/Azure AI Foundry): Project contains consent, datasets, models, endpoints. Microsoft Learn

Add voice talent consent to the professional voice project: Upload the recorded consent statement for the voice talent. Microsoft Learn

Professional voice fine-tuning data – required formats: Individual utterances + transcript require WAV/RIFF PCM; guidance on sample rate and transcript format. Microsoft Learn

Add a professional voice training dataset & resolve data issues: Data validation, pronunciation/noise checks, and the Analyze data step prior to training. Microsoft Le arn

Train your professional voice model: Start training and select Neural method. Microsoft Learn

Azure AI Content Safety supports custom blocklists —dictionaries of terms/phrases you define to detect obscure or domain-specific offensive terms in user content. Creating a blocklist is the minimal-effort, built-in way to meet this requirement; you don’t need to build a classifier.

References

Create and use blocklists in Azure AI Content Safety.

Text moderation overview (using built-in classifiers with optional blocklists).

Comprehensive Detailed Explanation

Active learning in Language Understanding (LUIS / Conversational Language Understanding) works by collecting real user queries sent to your prediction endpoint.

By setting log=true in the prediction endpoint query, these user utterances are stored in the LUIS service.

The logged utterances can then be reviewed in the Review endpoint utterances page of the LUIS portal, where you can relabel and retrain the model. This enables the active learning loop.

Why not the others?

A. show-all-intents=true: This simply forces the endpoint to return all intents, not related to active learning.

B. Enable speech priming: Used for improving speech-to-text accuracy, unrelated to LUIS active learning.

D. Enable sentiment analysis: Adds sentiment scores to prediction output, not related to active learning.