AI-102 Language, Speech, and Conversation

Since airports can appear in many forms and combinations (Example variations:

JFK

John F Kennedy Airport

New York JFK

JFK Airport), Pattern.any looks better. List is used for Small, fixed vocabulary.

List entities ARE great for airports
BUT the question specifically says "similar utterances" - this is Pattern.any territory

A list entity is perfect for airport names and codes because they represent a finite, well-defined set of values with known synonyms. This approach dramatically reduces the number of training utterances needed by allowing you to generalize utterance patterns and handle all airport variations through the predefined list mappings rather than requiring separate training examples for each airport.

List entities allow you to define a set of known values (like airport names or codes) and their synonyms.
When an utterance contains any of these values, LUIS will recognize them without needing to add separate utterances for each airport.
This minimizes the number of training utterances because you don’t have to create variations for every airport name or code.

Not A (Pattern.any) because that entity just captures free text placeholders without restricting values.
You’d still need many utterances to cover variations.

With D (list), you define all airport names/codes once, and the model generalizes across intents—minimizing training utterances

Pattern.any entities are ideal when you want to reduce the number of utterances needed for training.

They act as placeholders in patterns (like {fromAirport} or {toAirport}), and can match any word or phrase, such as airport names or codes.

This is especially useful when you have many similar utterances that differ only by variable data like city names, airport codes, etc.

To minimize the number of utterances needed for training when many intents share similar utterances with variable airport names or codes, you should use the Pattern.any custom entity. This entity type allows you to define patterns with placeholders for variable parts (like airport names/codes), reducing the need to enumerate every possible utterance variation.

Suppose you have intents like:
BookFlight
CheckFlightStatus

And users might say things like:
"Book a flight to JFK"
"Book a flight to London Heathrow"
"Check the status of flight DL123"
"Check the status of flight to SFO"

The variable parts are airport names, airport codes, or flight numbers.

You define a Pattern.any entity called, for example, @destination or @flightCode.

Intent: BookFlight
Pattern:
"Book a flight to {@destination}"
Intent: CheckFlightStatus
Pattern:
"Check the status of flight {@flightCode}"
"Check the status of flight to {@destination}"

Either A or D, both works. A less development effort, D a better solution. However, the requirement is not talking about minimizing development efforts.

Either A or D, both works. A less development effort, D a better solution. Howeve,r the requirements it is not talking to minimize development effort.

pattern.any https://learn.microsoft.com/ja-jp/azure/ai-services/luis/reference-entity-pattern-any?tabs=V2
list https://learn.microsoft.com/en-us/azure/ai-services/language-service/conversational-language-understanding/concepts/entity-components

List: fixed, close set of related words

It shoul be LIST

D. List
https://learn.microsoft.com/en-us/azure/ai-services/luis/reference-entity-list?tabs=V2
Suppose the app has a list, named Cities, allowing for variations of city names including city of airport (Sea-tac), airport code (SEA), postal zip code (98101), and phone area code (206).

List item Item synonyms
Seattle sea-tac, sea, 98101, 206, +1
Paris cdg, roissy, ory, 75001, 1, +33
book 2 tickets to paris

In the previous utterance, the word paris is mapped to the paris item as part of the Cities list entity. The list entity matches both the item's normalized name as well as the item synonyms.

LIST entity

USing a list entity allows you to define a set of values and their synonyms which will help minimize the number of utterances needed to train the model.

Hence the answer must be:
D) List

Pattern, you can use this to summarise multiple utterances into intents

"Patterns are designed to improve accuracy when multiple utterances are very similar. A pattern allows you to gain more accuracy for an intent without providing several more utterances."
So, it has to be pattern. The only one in the list is Pattern.Any.
Looks like the MIcrosoft guy just read this line and made the question and randomly chose a pattern type

The question is about the profile, not data.

A voice talent profile requires proof of consent from the person whose voice will be used. This ensures compliance with responsible AI policies.

A. a .zip file that contains 10-second .wav files and the associated transcripts as .txt files

Explanation:
This option aligns with the typical requirements for creating a custom neural voice. The system needs short, segmented audio samples along with their corresponding transcripts to accurately train the model and generate the synthetic voice.

By providing multiple short audio files and their transcripts, you ensure that the voice model can learn the nuances of the voice talent's speech and generate a high-quality custom voice.

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/record-custom-voice-samples

uploaded testing data must be A collection (.zip) of either .wave +.txt (if samples and transcripts are available)or .wave/.mp3 (if samples only available)
in both cases must be .zip

anwser is C, see here - https://learn.microsoft.com/en-us/azure/ai-services/speech-service/professional-voice-create-consent?pivots=speech-studio

Both me and ChatGPT agree with A

a .zip file that contains 10-second .wav files and the associated transcripts as .txt files

I say this answer is C.

C is answer.

C is the correct answer

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-voice-talent

Based on the Azure AI documentation, the correct option for creating a voice talent profile for a custom neural voice is:

C. a .wav or .mp3 file of the voice talent consenting to the creation of a synthetic version of their voice.

This is because the documentation specifies the need for a recording of the voice talent's consent statement, acknowledging the use of their voice recordings by a specified company to create and use a synthetic version of their voice

C is the correct answer as only wav & mp3 formats are allowed. Zip is not allowed.

I initially thought A, but now think it is C
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-voice-talent

zip is not allowed. only wav & mp3.
And made synthetic. =C
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-voice-talent
follow the pictures very carefully.

In Q#13 (How should you upload the samples?), the selected answer as option B (for speech samples to be used in training) corresponds to this Q#43’s option A.

Q#43 (What should you upload to the profile?) though, refers specifically to the voice talent’s profile, that should contain an audio sample of the given consent, and hereto “Only .wav and .mp3 files are accepted” (at the bottom of the screenshot provided in the link).
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-voice-talent#add-voice-talent

first drop-down should instead be: api-nam.cognitive.microsofttranslator.com this is because the case study specifically states under Business Requirements "Data storage and processing must occur in datacenters located in the United States."

see reference documentation for base urls per geo region:

https://docs.microsoft.com/en-us/azure/cognitive-services/translator/reference/v3-0-reference

From that wall of text who has the eye sight to pick "united states".

1. api-nam.cognitive.microsofttranslator.com
2. /translate

Is this question still available on May 21, 2024?

1. api-nam.cognitive.microsofttranslator.com
2. /translate

A modified version of this was in the exam today.

Does the exam for AI 102 includes labs or its multiple choice only.

api-nam
translate

in the "existing environment" there is
An Azure Cognitive Services resource named wwics
so it should be "resource name".cognitiveservices.azure.com : wwics.cognitiveservices.azure.com/translator/
Ref. https://docs.microsoft.com/en-us/azure/cognitive-services/translator/reference/v3-0-reference

The correct package for working with Azure AI service for text intent identification, like Model1, is:

A. azure-cognitiveservices-language-textanalytics

Therefore, you should add the azure-cognitiveservices-language-textanalytics package to App1 for configuring it to use Model1. This package provides functionalities for working with the Language API in Azure Cognitive Services, which includes text analytics capabilities, such as identifying language and sentiment, and can be used for processing text input to determine intent.

answer is B.
"Conversation App: It's used in extracting intents and entities in conversations"
- https://learn.microsoft.com/ko-kr/python/api/overview/azure/ai-language-conversations-readme?view=azure-python

"The (Text Analytics) API can be used to analyze unstructured text for tasks such as sentiment analysis, key phrase extraction and language detection."
https://learn.microsoft.com/en-us/python/api/azure-cognitiveservices-language-textanalytics/azure.cognitiveservices.language.textanalytics.textanalyticsclient?view=azure-python-previous

The textanalytics package for Python lists its capabilities, which don't include intent (https://learn.microsoft.com/en-us/python/api/overview/azure/ai-textanalytics-readme?view=azure-python); on the other hand, the converstationapp for python does reference intent (https://learn.microsoft.com/en-us/python/api/overview/azure/ai-language-conversations-readme?view=azure-python).

Option A, azure-cognitiveservices-language-textanalytics, is focused on text analytics capabilities such as sentiment analysis, key phrase extraction, and language detection. While it provides valuable tools for text analysis, it doesn't specifically cater to intent recognition within conversational contexts, which is what you're looking for in Model1.

On the other hand, Option B, azure-ai-language-conversations, is designed specifically for handling conversational AI tasks, including intent recognition. This makes it the most appropriate package for configuring App1 to use Model1 for identifying the intent of text input.

A is correct

What an e v i l question from Microsoft. It depends on the text.

Should be B:
this package is specifically designed for conversational language understanding, which includes intent recognition.

Selected Answer: B

I say this answer is B.

azure-cognitiveservices-speech.

B is answer.

B is right answer.

When identifying text input intent, it is usually aggregated to cognitiveservices. not just Microsoft, but mega cloud products and solutions have too many similar names and should be aggregated.

B is correct

Must be B

B. azure-ai-language-conversations
This package is part of the Azure AI SDK for Python and contains functionality for building applications that can understand user intent using language models.

Should be B.
Source: https://pypi.org/project/azure-ai-language-conversations/
"Conversation App: It's used in extracting intents and entities in conversations"

1. Create an Azure Machine Learning workspace.
2. Create and train the model in the Azure Machine Learning studio.
3. Create an endpoint for the model.
4. Connect the custom skill to the endpoint.
5. Rerun the indexer to enrich the index.

https://learn.microsoft.com/en-us/training/modules/build-azure-machine-learn-custom-skill-for-azure-cognitive-search/03-enrich-search-index-use-model
1.Create an Azure Machine Learning workspace.
2.Create and train the model in the Azure Machine Learning studio.
3.Create an endpoint for the model.
4.Provision an Azure AI Services resource and obtain the endpoint.
5.Connect the custom skill to the endpoint.

According to the exercise at https://microsoftlearning.github.io/mslearn-knowledge-mining/Instructions/Exercises/02-search-skills.html#create-azure-resources, you have to both create the Search resource and run the indexer to incorporate the skill -- the two highest voted answers at this point are split on which step belongs in the answer. I suspect that the question author is assuming the search is already created and that you are adding the custom skill to it. That means the expected answer is:

Create an Azure Machine Learning workspace.
2. Create and train the model in the Azure Machine Learning studio.
3. Create an endpoint for the model.
4. Connect the custom skill to the endpoint.
5. Rerun the indexer to enrich the index.

To implement an Azure AI Search resource that uses a custom skill based on sentiment analysis, follow these actions in sequence:

Create an Azure Machine Learning workspace: Set up the environment where you can create and manage your machine learning experiments and models.

Create and train the model in the Azure Machine Learning studio: Develop and train your custom sentiment analysis model using the resources in your Azure Machine Learning workspace.

Create an endpoint for the model: Deploy your trained model and create an endpoint to access the model via REST API.

Connect the custom skill to the endpoint: Configure Azure AI Search to use your custom skill by connecting it to the model's endpoint.

Rerun the indexer to enrich the index: Update the search index by rerunning the indexer, which will now use the custom skill to analyze sentiment and enrich the data in the search index.

ChatGPT:
1- Provision an Azure AI Services resource and obtain the endpoint.
2- Create an Azure Machine Learning workspace.
3- Create and train the model in the Azure Machine Learning studio.
4- Create an endpoint for the model.
6- Connect the custom skill to the endpoint

1. Create an Azure Machine Learning workspace.
2. Create and train the model in the Azure Machine Learning studio.
3. Create an endpoint for the model.
4. Connect the custom skill to the endpoint.
5. Rerun the indexer to enrich the index.

Based on the detailed steps provided in the Microsoft Learn documentation, the sequence of actions indeed doesn't require provisioning a separate Azure AI Services resource. Instead, the custom skill is directly connected to the Azure Machine Learning model endpoint.

https://learn.microsoft.com/en-us/training/modules/build-azure-machine-learn-custom-skill-for-azure-cognitive-search/03-enrich-search-index-use-model

1. Create an Azure Machine Learning workspace.
2. Create and train the model in the Azure Machine Learning studio.
3. Create an endpoint for the model.
4. Connect the custom skill to the endpoint.
5. Rerun the indexer to enrich the index.

Create an Azure Machine Learning workspace.
Create and train the model in the Azure Machine Learning studio.
Create an endpoint for the model.
Provision an Azure AI Services resource and obtain the endpoint.
Connect the custom skill the endpoint.

Create an Azure Machine Learning workspace.
Create and train the model in the Azure Machine Learning studio.
Create an endpoint for the model.
Connect the custom skill to the endpoint.
Rerun the indexer to enrich the index.

Create workspace
Provision
Connect
Create studio
Rerun

Create an Azure Machine Learning workspace.
Create and train the model in the Azure Machine Learning studio.
Create an endpoint for the model.
Provision an Azure AI Services resource and obtain the endpoint.
Connect the custom skill to the endpoint.

Create an endpoint for the model.
Create and train the model in the Azure Machine Learning studio.
Create an Azure Machine Learning workspace.
Connect the custom skill to the endpoint.

so, with the options:
Create an Azure Machine Learning
Create and train
Provision an Azure
Create an endpoint
Connect

Create an Azure Machine Learning workspace.
Create and train the model in the Azure Machine Learning studio.
Create an endpoint for the model.
Provision an Azure AI Services resource and obtain the endpoint.
Connect the custom skill to the endpoint.

According to GPT4:

1. Create an Azure Machine Learning workspace.
2. Create and train the model in the Azure Machine Learning studio.
3. Provision an Azure AI Services resource and obtain the endpoint.
4. Create an endpoint for the model.
Connect the custom skill to the endpoint.
Rerun the indexer to enrich the index.

The answers should be
Resource Name
classify
ocp-Apim-Subscription-Key

autocorrect has nothing to do with the question

Except, of course, you need to set the Resource name and subscription key even if you're not doing moderation so it's really only classify that has to be set to enable moderation.

Resource Name
classify
ocp-Apim-Subscription-Key

NagaoShingo 1 month, 1 week ago
1. Resource name
2. classify
3. Ocp-Api-Suscription-Key

Resource Name: Provide the name of your Content Moderator resource.
Classify: Set to true to enable classification of content.
Ocp-Apim-Subscription-Key: Provide your subscription key for authentication.

All other fields are not necessary to set.

1. Resource name
2. classify
3. Ocp-Api-Suscription-Key

Resource name
classify
Ocp-Api-Suscription-Key

1. name[language] 2. "$when": "${stockLevel} != 'OK'" 3. image.altText[language]

File JSON

{
"sku": "b1",
"name": {
"en": "Bicycle",
"es": "Bicicleta",
"pt": "Bicicleta"
},
"stocklevel": "Out of Stock",
"description": {
"en": "Bicycle",
"es": "Bicicleta",
"pt": "Bicicleta"
},
"image": {
"uri": "https://upload.wordwideimporters.org/bicycle.jpg",
"alttext": {
"en": "Bicycle",
"es": "Bicicleta",
"pt": "Bicicleta"
}
},
"createdUtc": "2020-0214T06:08:39Z",
"language": "en"
}

1. name.en

2. "$when": "${stockLevel != 'OK'}"

3. image.altText.en

Let's evaluate why the other options for answer3 are not valid:
image.altText.en:
This option hardcodes the language to English (en).
image.altText.language:
This option assumes language is a property under altText. However, language is a variable representing the selected language (e.g., en, es, pt), not a property within altText.
image.altText.[“language”]:
This option incorrectly uses quotation marks around language, treating it as a literal string rather than a variable. This would attempt to look for an alt text property with the name "language", which does not exist.
Correct Option: image.altText.[language]
Explanation: This option correctly uses the language variable to dynamically access the appropriate alt text based on the user's selected language. It ensures that the alt text is in the correct language (English, Spanish, or Portuguese) by referencing the variable directly. This fulfills the requirement for a multilingual customer experience.

1. name.en

2. "$when": "${stockLevel != 'OK'}"

3. image.altText.en

Requirement: "Support interactions in English, Spanish, and Portuguese"

If you hard set it on english, how can you satisfy this requirement?
must be name[language] on top and AltText.[language] on bottom.

How I understood it:
The description in the first table says that the app and the user input is in english => app is in english => adaptive card should be in english
In addition to that, the provided JSON has a language property with value 'en'.
So, my guess is:
1. name.en
2. "$when": "${stockLevel != 'OK'}"
3. image.altText.en

1. name.en
2. "$when": "${stockLevel != 'OK'}"
3. image.altText.en

Is this question still available on May 21, 2024?

You don’t want code it in a way that the code just works with en - it’s a bad coding practice.

The given answers are correct.

The correct answer pairs for box1 and box3 could be-
name.en and image.altText.en
name["language"] and image.altText["language"]

since name["language"] does not exist as an option in box3, we will have to choose the first pair as the correct answer.

https://www.w3schools.com/js/js_json_syntax.asp

This is a repeat question, but this time the answer does not give an explanation or reference document. Is this answer correct?

it seems a repeat question, but with a slight different option. The earlier one with altText.[language], but this altText[language] without .
which one is right?

If there was an option for name.[language] instead of name[language] then I'd use that for more than one language option, otherwise we can't. For name[language], language would have to be an integer but it's a string, wouldn't work.
But.. for name.[language], that works.
https://stackoverflow.com/questions/882727/is-there-a-way-to-use-variable-keys-in-a-javascript-object-literal

No.
1.Export model1 to Host1.
2.Request approval to run the container.
3.Run the container.

The issue here, for me, is the order which is indeterminate. You do need to export the model and you do need to run the container (you do not need to retrain the model unless you change the domain). However, you can ask for permission to run the container either before or after exporting the model (though it makes sense to do that just before running the container). So, presumably Request/Export/Run and Export/Run/Request are both correct answers.

Export model1 to Host1
→ You need to export the custom model from Azure to your Docker host so it’s available locally.

Request approval to run the container
→ Azure Speech container deployment typically requires Microsoft approval (e.g., for speech containers using Custom Speech models).

Run the container
→ Once the model is exported and approval is granted, you can start the Docker container that runs the speech-to-text service

Request approval to run container
Export model1 to Host1
Run the container

Maybe the question creator was in a hurry (they always are) and saw only the documentation that said there needs to be approval (but it was for disconnected and even though this is not disconnected). Or maybe the question from ET missed disconnected keyword.

1. Request approval to run container
2. Export model1 to Host1
3. Run the container

this is the answer from Copilot:

To run the custom speech-to-text model on your Docker host, you should perform the following actions in sequence:

Export model to Host1: You need to export the model from Azure to your Docker host. This typically involves downloading the model and moving it to the Docker host.

Run the container: Once the model is on the Docker host, you can run the container that uses this model.

Configure disk logging: After the container is running, configure disk logging to keep track of the container's activities and to troubleshoot any issues that might arise.

Request
Export
Run

You only need to request approval if you plan to run the container in a completely disconnected environment. And you might not get approved at all as there are some requirements. The question does not indicate if we need to run the container in such a environment.

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-container-cstt?tabs=custom-model%2Ccontainer&pivots=programming-language-csharp
In this article, you'll learn how to download [from Microsoft Container Registry (MCR)], install, and run a Custom speech to text container.
Get the model ID (to use as the argument to the ModelId parameter of the docker run command): The custom model has to have been trained by using the Speech Studio.
N/a so far, excluding A (Retrain the model).

Not sure.
https://learn.microsoft.com/en-us/azure/cognitive-services/speech-service/speech-container-overview#request-approval-to-run-the-container

The only reasons I can think of for picking where the glossary file goes
a) The app has read permission for the German resource and write permission for the French resouce, so the glossary has to go in the German resource
b) The resource is being used as a proxy for the language the glossary targets. If so, this would need to be a German-to-French glossary -- see https://learn.microsoft.com/en-us/azure/ai-services/translator/document-translation/how-to-guides/create-use-glossaries -- and would belong with the German files.

The given answer is correct.

1. Upload a Glossary file to the french files container
2. Define a document translation specification that has french target
3. Perform asynchronous translation by using the document specification

This is the answer according to ChatGPT.
Why not Glossary file in German container? Glossary contains the specific terms for the TARGET language. French in this case. So it make sense to keep it in French file
container

I think that correct is the original response, Because the glossary file is to be in the source language. On the URL : https://learn.microsoft.com/en-us/azure/ai-services/translator/document-translation/how-to-guides/create-use-glossaries gives an example of english to french and in session
Create, upload, and use a glossary file says : ... "The following English-source glossary contains words that can have different meanings depending upon the context in which they're used". So the glossary needs to be in the source language and not in the target language. So the answer is
1. Upload a glossary file to the container for German files
2. Define a document translation specification with French as the target language
3. Perform an asynchronous translation by using the document translation specification
Ps.: I tested on chatGPT and the response goes to the same.

was on exam 20.06.24

curl -X POST "https://<your-region>.cognitiveservices.azure.com/translator/text/batch/v1.1/translator/document/batches?api-version=2024-05-01" \
-H "Ocp-Apim-Subscription-Key: <your-subscription-key>" \
-H "Content-Type: application/json" \
-d <insert JSON document here to specific source, target and glossary>

{
"inputs": [
{
"source": {
"sourceUrl": "https://<your-storage-account>.blob.core.windows.net/german-files",
"storageSource": "AzureBlob"
},
"targets": [
{
"targetUrl": "https://<your-storage-account>.blob.core.windows.net/french-files",
"storageSource": "AzureBlob",
"language": "fr",
"glossaries": [
{
"glossaryUrl": "https://<your-storage-account>.blob.core.windows.net/glossaries/de-fr-glossary.tsv",
"format": "TSV"
}
]
}
]
}
]
}

Well, reality is that in MS example glossary in own container. But it is also in "targets" section of json file, so answer "French container" probably safer.

1. Upload a Glossary file to the french files container
2. Define a document translation specification that has french target
3. Perform asynchronous translation by using the document specification

The given answer is correct : reason is below
Upload the terminology list: Begin by uploading a custom terminology list, essential for accurate translations. Ensure it's placed in a location recognized by the translation service.

Define translation specifications: Specify source (German), target (French), and any special parameters in a translation specification.

Execute asynchronous translation: Initiate the translation process based on the specification, which will preserve original formatting and save translated files in the French document container.

These three steps form the core of the solution, although practical implementation may require additional configuration and permissions. Ensure the terminology list is uploaded to a location compatible with the translation service's requirements.

You need a French Glossary in the French container - how else can the translator know the French Jargon

The glossary file can be in this own container and that’s tricky as in theory you can put it anywhere. But since we need probably need have different permissions for the source and the target container putting it in the container where the source files would do. Also pay attention to what a glossary file its content looks like and how to specify the location in a request sent to the endpoint. https://learn.microsoft.com/en-us/azure/ai-services/translator/document-translation/how-to-guides/create-use-glossaries

It needs to have internet connectivity to send billing information, so you cannot select an isolated K8s service.
1. K8s with internet
2. Image from MCR
3. AppID and Secret

1. Provision on-premise k8 cluster that is connected to Internet
2. Pull image from MCR
3. Run container and specify API key and endpoint URL of Cognitive Services Services

Internet Connectivity is required for billing

I don't think that internet connectivity is required either to send billing data or to access the service in Azure: You can do both over the VNet connection that your enterprise will have created to Azure. To quote Microsoft: "Docker must be configured to allow the containers to connect with and send billing data to Azure. ailure to allowlist various network channels that the Azure AI containers rely on will prevent the containe
r from working" which doesn't mean internet connectivity.

first of all you need to provision cluster , after in a pod or deployment in containers section tag image specify the image to pull from MCR and Run container and specify API key and endpoint URL of Cognitive Services Services , cluster needs internet access to can reach that endpoint.

k8s with internet : This is necessary because you still need to reach Azure Cognitive Services, which is hosted on Azure and thus requires internet connectivity.

1. provision K8 with internet
2. pull from MCR
3. Run container and specify API key and endpoint URL of Cognitive Services Services,

1. Provision on-premise k8 cluster that is isolated from Internet
2. Pull image from MCR
3. Run container and specify API key and endpoint URL of Cognitive Services Services

1. Provision an on-premises Kubernetes cluster that is isolated from the internet.
2. Pull an image from the Microsoft Container Registry (MCR).
3. Run the container and specify an API key and the Endpoint URL of the Cognitive Services resource.

Final Answer:
1. Provision on-premise k8 cluster that is isolated from Internet
2. Pull image from MCR
3. Run container and specify API key and endpoint URL of Cognitive Services Services

The Azure AI containers are required to submit metering information for billing purposes. Failure to allowlist various network channels that the Azure AI containers rely on will prevent the container from working.

https://learn.microsoft.com/en-us/azure/ai-services/cognitive-services-container-support

This was in 2-Mar-24 exam

The Image needs to be able to send billing info to the Azure AI Service. Also the question does not say anything about no internet traffic. It only states that the docs should remain on-premise.


Build a K8s cluster onprem with internet acecss.
Pull an image
specify the API key and URL

Request access to use containers in disconnected environments might not get approved what then? A on-premise k8 cluster with internet connectivity is the way to go for billing purpose. The documents still remain on premise.

if your cluster is isolated from Internet, how can you pull image from MCR?

1. Provision an on-premises Kubernetes cluster that is isolated from the internet.
2. Pull an image from MCR (Microsoft Container Registry) or your private container registry.
3. Deploy the app container on the on-premises Kubernetes cluster, specifying the API Key and Endpoint URL of the Cognitive Services resource.

FROM CHATGPT

- Provision an on-prem kubernetes cluster that is isolated from Internet
- Pull an image from MCR
- Run the container and specify an API Key and Endpoint URL of the Cognitive Services resource

It is true.

Should by NYN.
Key phrases would be (tested with API): Eiffel Tower, tour, Paris, visit
Entities are: Eiffen Tower, Paris

Duplicate question. The answer should be NYN.

I ran this in my local, confirmed this is the output,

Key Words:
Paris
Eiffel Tower

So it is N N N

NYN are right answer.

There is a separate function to print key phrase to the console so the answer is NO. They are tricky you into thinking that key phrase is the same thing as key word which its not

Final Answer:
N
Y
N

The correct answers are NYY. The last one is yes because it will send info to the console as per the code snippet

Same : Topic 3, Question 48. NYN.

To deploy a bot to Azure, at a minimum you need:

An app registration in Azure AD to represent the bot identity
An App Service instance to host the bot web service
An App Service plan which defines the pricing tier and scale for the App Service

We can not have an azure bot resource created without an App Registration.
Since the statement is: You create an azure bot (means the resource is already created; the app registration already created).

When you create an Azure Bot resource in the portal, part of what that resource does under the hood is register an application for you in Azure Active Directory. In other words, the Bot resource itself takes care of the AAD app registration automatically. Therefore, you do not need to create a separate app registration yourself—meaning option A includes an extra, unnecessary step.

A is the correct answer for this.

A is answer.

No doubt is A

App registry is only required for OnPremise App
https://learn.microsoft.com/en-us/azure/bot-service/bot-service-quickstart-registration?view=azure-bot-service-4.0&tabs=userassigned

Entra ID is surplus to requirement surely.

Therefore, the correct answer is A. only an app registration in Microsoft Azure Active Directory (Azure AD), part of Microsoft Entra, an Azure App Service instance, and an App Service plan. These components work together to ensure your bot is up and running in Azure.

A is correct

https://learn.microsoft.com/en-us/azure/bot-service/provision-and-publish-a-bot?view=azure-bot-service-4.0&tabs=userassigned%2Ccsharp

Your bot will use these types of resources.

The Azure subscription that you'll use to provision, publish, and manage the bot
One or more resource groups
A user-assigned managed identity or an Microsoft Entra ID app registration
An App Service Plan resource
An App Service resource
An Azure Bot resource

AD it's not necessary, aks doesn't have sense here

I think the correct answer should be C. An Azure Service Plan and an Azure Service instance should are necessary to deploy the bot to Azure.

The answer is wrong.
Correct answer should be:
Box1:
https://eastus.api.cognitive.microsoft.com

Box2:
/text/analytics/v3.1/languages

REF:
https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/languages/languages?tabs=HTTP.

NOTE:
Pay special attention to the Sample Request provided. Request to the API should be of the form:

POST {Endpoint}/text/analytics/v3.0/languages

Where the {Endpoint} as stated under the sub-heading "URI Parameters" was described as quoted here (see "Description" column of the table):

"Supported Cognitive Services endpoints (protocol and hostname, for example: https://westus.api.cognitive.microsoft.com)."

So the sample given shows the correct format of the *endpoint* as https://{location}.api.cognitive.microsoft.com

We only need to detect the language, so the Language Cognitive Service could be used, i.e.:
POST {Endpoint}/text/analytics/v3.0/languages
where Endpoint could be for example "westus"
https://docs.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/languages/languages?tabs=HTTP

Microsoft Co-Pilot said,

✅ Correct Solution
You are building an app that will process incoming email/direct messages and route them to English or French support teams.
That means the app must:

Detect language → to know if the message is EN or FR
Translate to English or French → depending on routing needs

So you must use:
Base URL
✔ api.cognitive.microsofttranslator.com
This is the correct endpoint for Translator Service.
*****(Not EastUS Cognitive, not portal.azure.com)*****

1. eastus.api.cognitive.microsoft.com
2. /text/analytics/v3.1/languages

1. eastus.api.cognitive.microsoft.com
2. text/analytics/v3.1/languages

on exam today 24/05/2024, Adedoyin_Simeon's response is the one I selected as well.

eastus.api.cognitive.microsoft.com and /text/analytics/v3.1/languages

Final Answer:
Box1:
https://eastus.api.cognitive.microsoft.com
Box2:
/text/analytics/v3.1/languages

This was in 2-Mar-24 exam

https://eastus.dev.cognitive.microsoft.com/docs/services/TextAnalytics.V2.0/operations/56f30ceeeda5650db055a3c7

This came in exam 28/06/2023

The scenario needs a Language service. Specifically just a way to know if its in english or french. No translation is needed. Hence:
Box1:
https://eastus.api.cognitive.microsoft.com
Box2:
/text/analytics/v3.1/languages

Source: https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.1preview4/languages/languages?tabs=HTTP

api.cogitive.microsofttranslator.com
eastus.api.cognitive.microsoft.com
portal.azure.com
Out of these three options, only option 2 is a valid Azure Cognitive Services endpoint. The eastus.api.cognitive.microsoft.com endpoint is used for various Azure Cognitive Services, including the Text Analytics and Translator services.

Option 1, api.cogitive.microsofttranslator.com, is not a valid endpoint for Azure Cognitive Services. It appears to be a typo, as "cogitive" is misspelled.

Option 3, portal.azure.com, is the URL for the Azure portal, which is a web-based management interface for Azure services. It is not an endpoint for Azure Cognitive Services APIs.

-./text/analytics/v3.1/languages
One of the APIs that you can use to process incoming email and direct messages to either French or English language support teams is the Text Analytics API. The specific API endpoint you can use is /text/analytics/v3.1/languages. This API can detect the language of a given text and return the ISO 639-1 language code. You can then use this information to route the message to the appropriate language support team.

mk1967 and Adedoyin_Simeon are correct
The correct answer should be:
Box1:
https://eastus.api.cognitive.microsoft.com

Box2:
/text/analytics/v3.1/languages

The answer is wrong. Correct answer should be:
https://eastus.api.cognitive.microsoft.com
&
/text/analytics/v3.1/languages

REF:
https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/languages/languages?tabs=HTTP.

NOTE:
Pay special attention to the Sample Request provided. Request to the API should be of the form:

POST {Endpoint}/text/analytics/v3.0/languages

Where the {Endpoint} as stated under the sub-heading "URI Parameters" was described as quoted here (see "Description" column of the table):

"Supported Cognitive Services endpoints (protocol and hostname, for example: https://westus.api.cognitive.microsoft.com)."

So the sample given shows the correct format of the *endpoint* as https://{location}.api.cognitive.microsoft.com

The answer is wrong. Correct answer should be:
https://eastus.api.cognitive.microsoft.com
&
/text/analytics/v3.1/languages

REF:
https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/languages/languages?tabs=HTTP.

NOTE:
Pay special attention to the Sample Request provided. Request to the API should be of the form:

POST {Endpoint}/text/analytics/v3.0/languages

Where the {Endpoint} as stated under the sub-heading "URI Parameters" was described as quoted here

"Supported Cognitive Services endpoints (protocol and hostname, for example: https://westus.api.cognitive.microsoft.com)."

So the sample given shows the correct format of the *endpoint* as https://{location}.api.cognitive.microsoft.com

Answer is incorrect. Language Cognitive Service is sufficient.
https://learn.microsoft.com/en-us/rest/api/cognitiveservices-textanalytics/3.0/languages/languages?tabs=HTTP

It should be:
- Speech to Text with AutoDetectSourceLanguageConfig. It can't be Text Analytics because the input is callers' voice.
- Text to Speech: the output is voice.

Speech-to-Text : https://docs.microsoft.com/en-us/azure/cognitive-services/speech-service/speech-to-text

Text-to-Speech : https://docs.microsoft.com/en-us/azure/cognitive-services/speech-service/text-to-speech

Both support common languages, including French.

https://docs.microsoft.com/en-us/azure/cognitive-services/speech-service/language-support?tabs=speechtotext

✔ Detect the incoming language:
Translator

Translator includes built‑in language detection for text.
Even though the input is voice, the typical flow is: Speech‑to‑Text → Translator (detect language + translate).
Among the listed services, Translator is the only one capable of detecting language.

1. Speech to Text
2. Text to Speech

https://learn.microsoft.com/en-us/azure/synapse-analytics/machine-learning/tutorial-text-analytics-use-mmlspark

1. Speech to Text
2. Text to Speech

1. Speech to Text
2. Text to Speech

ChatGPT: speech to text + text to speech

The answer should be Text Analytics and TTS.

1. Speech to Text
2. Test to Speech

From Takedajuku perspective, if you study for 4 days and spend 2 days reviewing, you will have a better chance of passing the exam.

Final Answer:
1. Speech to Text
2. Text to Speech

Language identification (LID) use cases include:
Speech to text recognition when you need to identify the language in an audio source and then transcribe it to text.
Speech translation when you need to identify the language in an audio source and then translate it to another language.
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-identification?tabs=once&pivots=programming-language-python

People, read the descriptions, these are describing only parts of the process, not the whole process.
1) DETECT incoming language has got not to do with speech to text, but with text analytics. Yes, speech to text is a part of the whole process but this services TRANSCRIBES and does not do language detection.
2) Respond to caller's language should be Text to speech

It can't be Text Analytics. It is an incoming call. It should be Speech to Text
https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-identification?tabs=once&pivots=programming-language-csharp

And then Text to Speech for the response

Speech to text - as it's based on a voice call
Text to speech - it's not translator here, as the key thing is that the program needs to respond, not just translate

At this time, this question is out of date. See https://learn.microsoft.com/en-us/azure/ai-services/language-service/language-detection/overview . In old time, Text Analytics is correct.

to detect the incoming language you need Text analytics, Speech to text does not have an option to detect language

according to https://learn.microsoft.com/en-us/azure/ai-services/language-service/overview
Azure AI Language now unifies the Text Analyrics, QnA Maker and LUIS. So in my opinion, the question is obsolete.

without dispatch you select the manually decide when call to Qna and LUIS .. thats violates for automation. QnA Maker, Language Understanding, and Dispatch

D. Text Analytics, Translator, and Dispatch
Text Analytics analyzes user messages for sentiment and other insights
Translator handles language translation for multilingual support
Dispatch intelligently routes requests to the appropriate underlying services (knowledge base, chit-chat models, etc.) and selects the best model automatically

Dispatch is used to route the user's query to the appropriate model (LUIS, QnA Maker, or others).

🔸 Translator is needed for multilingual support.

LUIS → Understands user intent and supports multilingual.

Text Analytics → Does sentiment analysis.

QnA Maker → For chit-chat and FAQ knowledge base.

So Other options don’t meet all features. Only C checks all boxes

Option C: Language Understanding, Text Analytics, and QnA Maker
• Language Understanding = LUIS, handles intent detection.
• Text Analytics = handles sentiment analysis.
• QnA Maker = handles chit-chat and knowledge base.
This meets all three requirements.

Why not A?
Option A: QnA Maker, Language Understanding, and Dispatch
• QnA Maker handles knowledge base and chit-chat.
• Language Understanding (LUIS) enables intent and entity recognition.
• Dispatch routes to different services, including QnA or LUIS, based on user input.

BUT: No sentiment analysis capability included

C. Language Understanding, Text Analytics, and QnA Maker
It meets all the functional requirements:

Chitchat + KB → QnA Maker

Multilingual → LUIS and QnA both support it natively

Sentiment → Text Analytics

Model selection → handled by bot framework routing logic or Dispatch (not strictly required if you're only using two models)

Why the others don’t fully match:
A. QnA Maker, Language Understanding, and Dispatch
No sentiment analysis or multilingual support.
B. Translator, Speech, and Dispatch
No sentiment analysis or knowledge base.
C. Language Understanding, Text Analytics, and QnA Maker
Lacks multilingual capability and automatic model selection.

A: no sentiment analysis
B: don't need speech and no sentiment analysis
C: can't select the best language model automatically. -> no dispatch

The best choice here would be Option A: QnA Maker, Language Understanding, and Dispatch.

Here's why:
- QnA Maker supports the knowledge base requirement.
- Language Understanding (LUIS) supports chit-chat by interpreting user intents and entities.
- Dispatch helps in selecting the best language model automatically and routing the requests to the appropriate service.

The combination of these three components covers all the necessary requirements for your chatbot. Option B and D do not include the knowledge base component, and Option C lacks the automatic selection of the best language model.

Option A (QnA Maker, Language Understanding, and Dispatch): This combination supports chit-chat, knowledge base, multilingual models, and automatic language model selection through Dispatch.

Option C (Language Understanding, Text Analytics, and QnA Maker): This option includes Text Analytics for sentiment analysis but lacks Dispatch for automatic model selection.

Option D (Text Analytics, Translator, and Dispatch): This option includes Text Analytics and Dispatch but lacks QnA Maker for knowledge bases.

Given these considerations, Option A would be the most suitable choice as it covers all requirements through the combined capabilities of QnA Maker, Language Understanding, and Dispatch.

Language Understanding include Sentiment analysis.
And use the Bot framework Dispatch CLI tool to help build a process to work with both services(Language Understanding, QnA Maker).
So Answer is A
https://learn.microsoft.com/en-us/azure/ai-services/luis/luis-reference-prebuilt-sentiment?tabs=english

C is the correct answer

I think the answer option is A (see https://learn.microsoft.com/en-us/azure/ai-services/qnamaker/choose-natural-language-processing-service)

C is right answer.

C is right.

As Luis is outdated (with its Dispatch option), I must be C

The link provided mentions addresses under 'ML Entities with Structure'. Will be hard to identify all possible international addresses with RegEx.

A - MAchine Learned

in documentation
https://learn.microsoft.com/en-us/azure/cognitive-services/LUIS/concepts/entities#ml-entity-with-structure

- ML Entity with Structure
An ML entity can be composed of smaller sub-entities, each of which can have its own properties. For example, an Address entity could have the following structure:
Address: 4567 Main Street, NY, 98052, USA
Building Number: 4567
Street Name: Main Street
State: NY
Zip Code: 98052
Country: USA

For the multiple-choice exam question you showed earlier, the correct answer is D. Pattern.any, because the scenario only asks to capture billing addresses — not to break them down into sub-entities.

Should be pattern Any

Azure bot service no longer part of the exam, no need to revise this topic - https://trainingsupport.microsoft.com/en-us/mcp/forum/all/azure-ai102-chatbots-service/798c0cfa-3475-474e-b4ec-8ab7fc790e81#:~:text=Yes%2C%20Julian%20is%20correct%2C%20based,not%20appear%20in%20the%20exam.

For me e GPT 3.5: machine learned

I know you don't know what I'm talking about, but if you think as Crossroads leads you, the answer is A.

As per Gemini chatbot, the answer is B. Regex. The following is the explanation it gives:
Regex offers a powerful and efficient way to define patterns for capturing specific text formats, making it ideal for structured data like billing addresses. You can design a regex pattern to capture elements like street address, city, state, zip code, and potentially country code.

; and why the answer is not 'machine learned':
While machine learning can be powerful for complex entity recognition, it might be overkill for a well-defined structure like billing addresses. Training a machine learning model could be time-consuming and resource-intensive compared to a simpler solution.

It must be A.

on exam, I selected machine learned

Regex would be a good solution if the addresses were in a specific area (e.g, USA). But since that information is not given, you have to assume any type of address. So the correct answer in this case is A.

it should be A

A. machine learned

Another one to memorise. A quick trick:
Q sounds like a C, for Cognitive
n
A for App Service

Right answer : at the creation, we have to precise Azure Search an Azure Web App details.
"When you create a QnAMaker resource, you host the data in your own Azure subscription. Azure Search is used to index your data." & "When you create a QnAMaker resource, you host the runtime in your own Azure subscription. App Service is the compute engine that runs the QnA Maker queries for you."

The correct answers are C and D:

C. Azure Storage - Required for storing knowledge base data, configurations, and logs
D. Azure Cognitive Search - Required for indexing and searching through the knowledge base content
These two services are fundamental dependencies that QnA Maker automatically provisions to support its core functionality: storing knowledge bases (Storage) and enabling fast question matching (Cognitive Search).

When you provision the Azure AI Language – Question Answering feature (the successor to QnA Maker), the following resources are automatically created in the specified resource group:
Azure Cognitive Search - Used to index and query your knowledge base documents.
Essential for enabling semantic search and retrieval-based question answering.

Azure Storage - Stores the source documents (e.g., PDFs, Word files, web pages) that are ingested into your knowledge base.

The Azure SQL Database is used by QnA Maker to store the knowledge base data.
Azure Cognitive Search is used to index and search the QnA pairs efficiently.

DE are correct answer

QnAMaker was terminated

You can check directly in azure.portal by creating a Q&A Maker. Now besides D and E it creates an App Insights service but this is not in the quiestion options.

Apart from the above pricing, you also pay for resources consumed as described below. When you create a QnA Maker resource, you host the data and the runtime in your own Azure subscription. These are powered by Azure AI Search and App Service.

https://azure.microsoft.com/en-us/pricing/details/cognitive-services/qna-maker/

I see only Azure AI Search service being provided and the only correct answer.

REF: https://azure.microsoft.com/en-us/pricing/details/cognitive-services/qna-maker/

"When you create a QnA Maker resource, you host the data and the runtime in your own Azure subscription. These are powered by Azure AI Search and App Service."

QnA Maker is now retired

D and E for Perplexity AI and Claude. Sadly from portal this resource cannot be provisioned anymore

C and D for ChatGPT

DE are right answer.

DE is power.

D and E are right.

B is the answer
In Language Understanding (LUIS) classic, there are two separate resources:

Authoring resource – used to create, edit, train, and publish the language model

Prediction resource – used only at runtime to query the published model

To add more contributors who can build or modify the language model, you must grant permissions on the authoring resource.

It's B. Check with AI

I selected B, but it does not make sense. It is like there is nothing sadder than having a breakup with a lover.

B is right.

The best answer is B. the Access control (IAM) page for the authoring resources in the Azure portal., to add more contributors to a Language Understanding (classic) resource, we should use the Access control (IAM) page for the authoring resources in the Azure portal. This allows we to manage access and permissions for the resource at a granular level, including adding new contributors.