📘 Free AI-901 Sample Questions
You are using the Azure Speech SDK to develop a Python application that supports real-time spoken
conversations.
Which Azure Speech class should you use to configure the connection to the Azure Speech service?
A
AudioOutputConfig
B
SpeechSynthesizer
C
AudioConfig
D
AudioConfig
Correct Answer:
D. AudioConfig
Explanation:
Technical Justification for Correct Answer: C. input_image
For the given scenario involving a Microsoft Foundry project with a vision-enabled model deployment and the
use of the Azure OpenAI Responses API, the correct content item to include in the request for image analysis
is C. input_image. Here's why:
Correct Answer: C. input_image
Justification: When sending a prompt to a vision-enabled model for analysis, the image being analyzed is
considered the input to the model. The Azure OpenAI Responses API, like most AI model inference APIs,
expects the data to be processed (in this case, the image) to be provided as part of the input payload.
Technical Alignment: The term "input_image" directly aligns with API design principles where inputs are the data
being processed, and outputs are the results of that processing.
Incorrect Options with Justifications:
A. image_generation
image based on a text prompt, not analyzing one.
B. output_image
content item's purpose, not its encoding.
References
Why Incorrect: This option suggests the generation of an image, which is contrary to the requirement of
analyzing an existing image. Image generation would be relevant if the model were tasked with creating an
Why Incorrect: "Output_image" implies the image is the result of the model's processing, which is not the
case here. The image is the subject of the analysis, not the outcome.
D. image_base64
Why Incorrect: While it's technically possible that the image might be encoded in Base64 for transmission (to
be included in a JSON payload, for example), "image_base64" describes the encoding format of the image,
not the purpose or role of the image in the request (i.e., as input for analysis). The question asks for the
Key Takeaway: The question tests understanding of the role of data in API requests (input vs. output) rather
than data encoding formats.
1. Azure OpenAI API Documentation - https://azure.microsoft.com/en-us/services/cognitive
services/openai/ (Refer to API examples for input structures)
2. Microsoft Azure - API Design Principles - https://docs.microsoft.com/en-us/azure/architecture/api
design/principles (See guidelines on input/output handling)
HOTSPOT -
Select the answer that correctly completes the sentence.
A
Correct Answer:
A.
Explanation:
a base64-encoded image data.
Why the Other Options Are Incorrect:
The OpenAI Responses and Chat Completions APIs natively support two methods for passing visual content to
vision-enabled models (like gpt-4o):An accessible, direct HTTP URL pointing to the hosted image.The raw
image data embedded directly in the API payload text stream, which must be formatted as a base64-encoded
image data string using the Data URL schema (e.g., data:image/jpeg;base64,/9j/4AAQ...).
a CSV file attachment: CSV is a structured plaintext spreadsheet format used to represent tabular dataset
parameters; it cannot be parsed natively by vision models as an image object.
an MP4 video stream: Standard vision endpoints look for discrete static image frames. Passing a raw, multi
frame video stream container directly into standard vision API picture arrays will trigger a payload format
validation failure.
a shared access signature (SAS) token: A SAS token is a Microsoft Azure-specific security credential used to
grant delegated access to Azure Storage objects. OpenAI APIs do not parse Azure storage authentication
tokens natively to pull backend blobs.
You need to create an AI agent in Microsoft Foundry that follows a specific role and behavior when responding to Users.
What should you configure?
A
system instructions
B
temperature
C
tokens per minute (TPM)
D
max completion tokens
Correct Answer:
A. system instructions
Explanation:
Technical Justification for Correct Answer: A. System Instructions
To create an AI agent in Microsoft Foundry that adheres to a specific role and behavior when responding to
users, configuring System Instructions is the most appropriate choice. Here's why:
A. System Instructions: This option allows you to define the AI agent's role, context, and behavioral
guidelines directly. By specifying system instructions, you can dictate the agent's persona, its limitations, and
how it should interact with users, ensuring it follows the desired behavior and role consistently. This is
fundamental in shaping the agent's overall interaction strategy.
Why Other Options are Less Suitable:
B. Temperature: Adjusting the temperature setting influences the randomness and creativity of the AI's
responses. While it can affect the tone and variability of answers, it does not directly control the agent's role
or core behavioral aspects in relation to user interactions. Temperature is more about response generation
style than adherence to a specific role or behavior.
C. Tokens Per Minute (TPM): TPM is a throughput or performance metric that controls the rate at which the AI
processes or generates text. It impacts the speed of response generation but has no bearing on the role or
behavioral aspects of the AI agent's interactions. Configuring TPM ensures scalability or performance
requirements are met, not role-specific behaviors.
requirement for role and behavioral compliance.
References
D. Max Completion Tokens: This setting limits the maximum number of tokens (units of text) in a response.
While crucial for controlling response length and potentially indirectly influencing behavior by limiting
response complexity, it does not directly define the role or specific behavioral guidelines for the AI agent.
Conclusion: For defining a specific role and behavior in an AI agent within Microsoft Foundry, System
Instructions (A) is the correct and most direct configuration to implement, as it directly addresses the
1. Microsoft Azure Documentation - Configure AI Agent Behaviors: https://docs.microsoft.com/en
us/azure/cognitive-services/language-service/custom-chat/configure-agent-behaviors
2. Microsoft Foundry AI Agent Configuration Guide: https://learn.microsoft.com/en-us/azure/machine
learning/service/managed-notebooks??????-foundry'ai-agent-configuration (Note: As of my last
update, a more specific "Microsoft Foundry" link might not be publicly available due to the evolving
nature of Microsoft's services. For the most relevant and up-to-date information, searching the
official Microsoft Azure or Microsoft Learn platforms with keywords like "Microsoft Foundry AI Agent
Configuration" is recommended. If a direct link is required for study purposes, consider the first link
for general AI configuration principles in Azure, which can be adapted to Foundry contexts.)
Actual Working Link Provided for General Azure AI Configuration (as direct Foundry docs might not be
publicly available or named slightly differently)
HOTSPOT -
For each of the following statements, select Yes if the statement is true. Otherwise, select No.
NOTE: Each correct selection is worth one point.
A
Correct Answer:
A.
Explanation:
System prompts can be used to authorize users. (No)
Authorization is an identity access management (IAM) security control handled by application code, API
gateways, or OAuth tokens. A system prompt is simply text input given to a language model and completely
lacks the capability to securely authenticate or authorize human users.
A system prompt is used to reduce tokens per minute (TPM). (No)
Tokens Per Minute (TPM) is a fixed infrastructural rate limit applied to an API endpoint by the platform
provider to manage compute usage. System prompts define the model's persona or operational boundaries;
they do not dictate or adjust the backend server's rate-limiting ceilings.
A system prompt guides the behavior of a generative Al model. (Yes)
The primary purpose of a system prompt (or system message) is to establish instructions, persona guidelines,
tone rules, constraints, and structural boundaries that direct exactly how the generative Al model behaves
and responds during user interaction.
You are developing a web app that processes invoices to calculate expenses.
You need to extract structured fields, including nested values, from the invoices by using a defined schema.
What should you use?
A
an optical character recognition (OCR)-only document processing pipeline
B
a transcription workflow in Azure Speech in Foundry Tools
C
an analyzer in Azure Content Understanding in Foundry Tools
D
an Azure AI Search service
Correct Answer:
C. an analyzer in Azure Content Understanding in Foundry Tools
Explanation:
Technical Justification: Choosing the Correct Option
Correct Answer: C. an analyzer in Azure Content Understanding in Foundry Tools
Why C is the best choice: Azure Content Understanding in Foundry Tools is specifically designed for
extracting structured information from unstructured or semi-structured data sources, such as invoices. Its
analyzers can be configured with a defined schema to recognize and extract both simple and nested values
from documents. This capability aligns perfectly with the requirement to process invoices based on a
predefined schema, making it the most suitable choice for this task.
Why Other Options are Less Suitable:
A. an optical character recognition (OCR)-only document processing pipeline
Limitation: While OCR is excellent for extracting text from images of documents, an OCR-only pipeline lacks
the built-in capability to understand the structure or semantics of the extracted text based on a predefined
schema. It would require additional, custom development for schema-based extraction, making it less
efficient and more resource-intensive compared to a purpose-built solution like Azure Content Understanding.
B. a transcription workflow in Azure Speech in Foundry Tools
Misalignment: Azure Speech is primarily designed for audio-to-text transcription and speech-related tasks. It
is not optimized for processing written documents like invoices or for extracting structured fields based on a
schema, making it a mismatch for this specific requirement.
D. an Azure Al Search service
Primary Function: Azure Al Search (formerly Azure Cognitive Search) is mainly used for creating searchable
indexes from various data sources. While it can process documents, its primary function is search capability
rather than extracting structured fields from documents based on a predefined schema, which is the core
requirement here.
References:
Azure Content Understanding Documentation - Explains how to use analyzers for structured data extraction.
Azure Cognitive Services Comparison - Helps in understanding the primary use cases for each service,
including Azure Speech, Azure Content Understanding, and Azure Al Search.
HOTSPOT -
For each of the following statements, select Yes if the statement is true. Otherwise, select No.
NOTE: Each correct selection is worth one point.
A
Correct Answer:
A.
Explanation:
Voice Live returns only transcribed text. (No)
Why it is No: The Azure Al Foundry Voice Live API is a comprehensive audio engine that functions as an end-
to-end speech-to-speech solution. Rather than outputting flat digital text lines, it delivers live streaming
synthesized audio tracks, real-time facial/lip-sync markers for interactive Al avatar configurations, and
execution metadata.
Voice Live requires you to separately implement speech to text and text to speech services. (No)
Why it is No: One of the primary advantages of this API is that it removes traditional pipeline fragmentation.
Historically, engineering a live voice agent required complex middleware code to connect an independent
Speech-to-Text tool, an LLM reasoning block, and a Text-to-Speech library. Voice Live is fully managed and
natively eliminates that integration overhead.
Voice Live combines speech to text, reasoning, and text to speech into a single conversational experience.
(Yes)
Why it is Yes: As confirmed in the Microsoft Learn Voice Live documentation, the service unifies speech
recognition, large language model generative reasoning, and low-latency vocal synthesis into a single,
streamlined backend pipeline endpoint optimized for conversational Al.
HOTSPOT -
You are developing a voice application that listens for spoken commands and converts them into text by using
Azure Speech in Foundry Tools.
How should you complete the Python code? To answer, select the appropriate option in the answer area.
NOTE: Each correct selection is worth one point.
A
Correct Answer:
A.
Explanation:
recognizer.recognize_once()
Why the Other Options Are Incorrect:
recognizer.speak_text_async("Ready"
recognizer.start_continuous_recognition()
In the Azure Speech SDK, the basic SpeechRecognizer object is instantiated to perform Speech-to-Text
(transcription). The standard method used to trigger a single-shot, short utterance recognition session (which
captures audio up to 15 seconds or until the first pause/silence is detected) is recognize_once(). This is the
foundational method used in almost all introductory, basic Azure Speech-to-Text code examples.
)Why it is incorrect: This method belongs to the SpeechSynthesizer class, which is used for Text-to-Speech
(TTS) operations. Calling a speech synthesis command on a SpeechRecognizer object will throw an
AttributeError runtime exception, as a speech recognition object cannot generate audio playback.
Why it is incorrect: While this is a valid method for long-form speech transcription (such as transcribing a
lecture, meeting, or long audio file), it operates purely asynchronously via event handlers (recognizing,
recognized, session_stopped). Using it requires setting up specific callback listener functions beforehand to
capture the streaming text data, which is absent from this simple linear code block.
recognizer.start_keyword_recognition()
Why it is incorrect: This method is designed specifically for wake-word or ambient keyword detection (e.g.,
listening continuously for a phrase like "Hey Cortana" or "Computer"). It requires passing a pre-compiled
keyword model file object (KeywordRecognitionModel) as an input parameter into the function, which is not
configured in the code baseline above.
What are two purposes of instructions when prompting a generative AI model? Each correct answer presents part
of the solution.
NOTE: Each correct selection is worth one point.
A
defines the agent’s role and behavior
B
selects which model to use
C
defines the Azure region where inference occurs
D
defines the tokens per minute (TPM) allocation for the model
E
defines constraints on the model’s responses
Correct Answer:
A. defines the agent’s role and behavior
Explanation:
Technical Justification for Correct Answer: AE
A. defines the agent's role and behaviorThis is a correct purpose of instructions when prompling a generative
Al model. By deining the agent's role and behavior, instructions guide the model on how to interact with the
user or the envionment from a specitc persona or operational contest. This ensures the model's output
aligns with the expected character, tone, and operational boundaries of the defined agent, enhancing the
coherence and relevance of the responses.
Why it's correat: Aligns model interaction with predefined operational and contextual perameters.
Technical Rationale: This approach is fundamental in conversstional Al and role-playing scenarics, where the
model must emulate specilic behaviors or personas.
E. defines constraints on the model's responsesThis is also a correct purpose. Defining constraints ensures
the model's output adheres to specifie guidelines, limitations, or formats required by the application er user.
Constraints can include content filters, response length, or adherence to certain themes or topics, thereby
managing the model's croativity within predeaned bounds.
Why it's correct: Ensures model outputs meet specific requirements or guidelines.
Technieal Rationale: Constrsint delniion is crucial for maintaining control over the model's output in
production environments, especiaily where compliance or salety is a concem.
Why Other Options are Less Suitable:
B. selects which modal to use
Incorrect because: The selection of a model to use is typically delermined before prompling the model, as
part ol the deplayment oe coniguration process, not through the prompt inatructions thersselves. Prompts
are used to guide the selected model's output, not to choose between models.
C. defines the Azure region where inference occurs
Incorrect because: The Azure region for inlerence is a deployment and infrastructure setting, configured
when seiting up the Al service on Microsoft Azure, not something defined through model prompts. Prompts
influence model behavior, not deployment geography.
D. defines the tokens per minute (TPM) allocation for the model
Incorrect because: TPM alocation is a resouico managemon and biling parameter, sot during the sbrvice
configuration to control throughput, not through individuall prompts to the model. Prompts do not dictate
operational capacities.
Refarences
1. Microsoft Azure Documentation - Designing Effective Prompts for Azure Cegnitive Services
Language Models
URL: Ittpad/docs.mierosolt.com/en-us/azureleognitive-services/language-service/language-
models/design-effective-prompts
Relevance: Provides guidance on crafting prompts, including seting roles and constraint.
2 Azure Cognitive Services . Best Practices for Working with Large Language Models
URL: httpsi/does.mierosolt.com/en-us'azuraleognitive-servicas/ang.ge-servicalarge-language-
models/best-practices
Relevance: Discusses the importance of consttsined prompting for controlied outputs.
HOTSPOT -
You are developing an application that converts text into spoken audio and saves the synthesized audio to a file by
using Azure Speech in Foundry Tools.
How should you complete the Python code? To answer, select the appropriate option in the answer area.
NOTE: Each correct selection is worth one point.
A
Correct Answer:
A.
Explanation:
Explanation:
AudioOutputConfig(filename="output.wav")
This question focuses on using the Azure AI Speech SDK for Python to perform Text-to-Speech (speech
synthesis) operations and save the output directly to a local audio file.When instantiating a
class constructor and specify the target string location using the filename named parameter.
Why the Other Options Are Incorrect:
AudioOutputConfig(stream.
syntactically invalid.
SpeechSynthesizer object, you can pass an audio_config parameter to control where the generated audio
stream is routed. To write the audio data directly to a local file container, you invoke the AudioOutputConfig
)Why it is incorrect: While AudioOutputConfig can write to custom push or pull memory streams, the
initialization parameter requires a properly configured stream instance (such as a PullAudioOutputStream
object). Simply passing an uninitialized or generic reference variable like stream in this basic context is
AudioStreamFormat(wave_stream_format=AudioStreamWaveFormat.PCM)
Why it is incorrect: The AudioStreamFormat class is used to describe the digital encoding properties, sample
rates, and bit depths of raw custom input/output streams. It is an informational layout descriptor rather than a
valid target routing configuration object that can be passed directly as the audio_config parameter to a
SpeechSynthesizer.
HOTSPOT -
Select the answer that correctly completes the sentence.
A
Correct Answer:
A.
Explanation:
you must configure a: system prompt.
follow, and how it handles user interactions.
Why the Other Options Are Incorrect:
do with AI behavioral logic.
A system prompt (or system message) is the core operational instruction given to a generative AI model or
autonomous agent to establish its persona, boundaries, tone, and step-by-step logic. It acts as the
foundational guide that explicitly dictates how the agent should behave, what knowledge constraints it must
deployment slot: This is an Azure App Service infrastructure feature used to host different versions of a web
application (e.g., staging vs. production) for seamless blue-green deployments and testing. It has nothing to
embedding index: This is a vectorized database repository used in Retrieval-Augmented Generation (RAG)
workflows to search and fetch relevant semantic document chunks. It grounds the agent with external facts,
but it does not shape its tone or behavior.
fine-tuning job: This is an offline machine learning training process used to permanently adjust the internal
weights of a base model using a specific specialized dataset. While it adapts a model to domain styles, an
agent's direct operational role and execution rules are configured dynamically using a system prompt.
Questions: 1-10 out of 52
Continue Full Practice..
GET ALL 52 QUESTIONS