Use Azure Speech and Language Understanding Services

https://docs.microsoft.com/en-us/learn/modules/use-language-understanding-speech/4-exercise-use-speech-services

You can integrate the Speech service with the Language Understanding service to create applications that can intelligently determine user intents from spoken input.

Note: This exercise works best if you have a microphone. Some hosted virtual environments may be able to capture audio from your local microphone, but if this doesn't work (or you don't have a microphone at all), you can use a provided audio file for speech input. Follow the instructions carefully, as you'll need to choose different options depending on whether you are using a microphone or the audio file.

Clone the repository for this course

If you have already cloned AI-102-AIEngineer code repository to the environment where you're working on this lab, open it in Visual Studio Code; otherwise, follow these steps to clone it now.

1. Start Visual Studio Code.

2. Open the palette (SHIFT+CTRL+P) and run a Git: Clone command to clone the https://github.com/MicrosoftLearning/AI-102-AIEngineer repository to a local folder (it doesn't matter which folder).

3. When the repository has been cloned, open the folder in Visual Studio Code.

4. Wait while additional files are installed to support the C# code projects in the repo.

   Note: If you are prompted to add required assets to build and debug, select Not Now.

Create Language Understanding resources

If you already have Language Understanding authoring and prediction resources in your Azure subscription, you can use them in this exercise. Otherwise, follow these instructions to create them.

1. Open the Azure portal at https://portal.azure.com, and sign in using the Microsoft account associated with your Azure subscription.

2. Select the ＋Create a resource button, search for language understanding, and create a Language Understanding resource with the following settings:

   • Create option: Both
   • Subscription: Your Azure subscription
   • Resource group: Choose or create a resource group (if you are using a restricted subscription, you may not have permission to create a new resource group - use the one provided)
   • Name: Enter a unique name
   • Authoring location: Select your preferred location
   • Authoring pricing tier: F0
   • Prediction location: Choose the same location as your authoring location
   • Prediction pricing tier: F0 (If F0 is not available, choose S0)

3. Wait for the resources to be created, and note that two Language Understanding resources are provisioned; one for authoring, and another for prediction. You can view both of these by navigating to the resource group where you created them.

Prepare a Language Understanding app

If you already have a Clock app from a previous exercise, open it in the Language Understanding portal at https://www.luis.ai. Otherwise, follow these instructions to create it.

1. In a new browser tab, open the Language Understanding portal at https://www.luis.ai.
2. Sign in using the Microsoft account associated with your Azure subscription. If this is the first time you have signed into the Language Understanding portal, you may need to grant the app some permissions to access your account details. Then complete the Welcome steps by selecting your Azure subscription and the authoring resource you just created.
3. Open the Conversation Apps page, next to ＋New app, view the drop-down list and select Import As LU. Browse to the 11-luis-speech subfolder in the project folder containing the lab files for this exercise, and select Clock.lu. Then specify a unique name for the clock app.
4. If a panel with tips for creating an effective Language Understanding app is displayed, close it.

Train and publish the app with Speech Priming

1. At the top of the Language Understanding portal, select Train to train the app if it has not already been trained.
2. At the top right of the Language Understanding portal, select Publish. Then select the Production slot and change the settings to enable Speech Priming (this will result in better performance for speech recognition).
3. After publishing is complete, at the top of the Language Understanding portal, select Manage.
4. On the Settings page, note the App ID. Client applications need this to use your app.
5. On the Azure Resources page, under Prediction resources, if no prediction resource is listed, add the prediction resource in your Azure subscription.
6. Note the Primary Key, Secondary Key, and Location (not endpoint!) for the prediction resource. Speech SDK client applications need the location and one of the keys to connect to the prediction resource and be authenticated.

Configure a client application for Language Understanding

In this exercise, you'll create a client application that accepts spoken input and uses your Language Understanding app to predict the user's intent.

Note: You can choose to use the SDK for either C# or Python in this exercise. In the steps that follow, perform the actions appropriate for your preferred language.

1. In Visual Studio Code, in the Explorer pane, browse to the 11-luis-speech folder and expand the C-Sharp or Python folder depending on your language preference.

2. View the contents of the speaking-clock-client folder, and note that it contains a file for configuration settings:

   • C#: appsettings.json
   • Python: .env

   Open the configuration file and update the configuration values it contains to include the App ID for your Language Understanding app, and the Location (not the full endpoint - for example, eastus) and one of the Keys for its prediction resource (from the Manage page for your app in the Language Understanding portal).

Install SDK Packages

To use the Speech SDK with the Language Understanding service, you need to install the Speech SDK package for your programming language.

1. In Visual Studio, right-click the speaking-clock-client folder and open an integrated terminal. Then install the Language Understanding SDK package by running the appropriate command for your language preference:

   C#

   dotnet add package Microsoft.CognitiveServices.Speech --version 1.14.0

   Python

   pip install azure-cognitiveservices-speech==1.14.0

2. Additionally, if your system does not have a working microphone, you will need to use an audio file to provide spoken input for your application. In this case, use the following commands to install an additional package so your program can play the audio file (you can skip this if you intend to use a microphone):

   C#

   dotnet add package System.Windows.Extensions --version 4.6.0 

   Python

   pip install playsound==1.2.2

3. Note that the speaking-clock-client folder contains a code file for the client application:

   • C#: Program.cs
   • Python: speaking-clock-client.py

4. Open the code file and at the top, under the existing namespace references, find the comment Import namespaces. Then, under this comment, add the following language-specific code to import the namespaces you will need to use the Speech SDK:

   C#

   C#

   // Import namespaces using Microsoft.CognitiveServices.Speech; using Microsoft.CognitiveServices.Speech.Audio; using Microsoft.CognitiveServices.Speech.Intent;

   Python

   Python

   # Import namespaces import azure.cognitiveservices.speech as speech_sdk

5. Additionally, if your system does not have a working microphone, under the existing namespace imports, add the following code to import the library you will use to play an audio file:

   C#

   C#

   using System.Media;

   Python

   Python

   from playsound import playsound

Create an IntentRecognizer

The IntentRecognizer class provides a client object that you can use to get Language Understanding predictions from spoken input.

1. In the Main function, note that code to load the App ID, prediction region, and key from the configuration file has already been provided. Then find the comment Configure speech service and get intent recognizer, and add the following code depending on whether you will use a microphone or an audio file for speech input:

   If you have a working microphone:

   C#

   C#

   // Configure speech service and get intent recognizer SpeechConfig speechConfig = SpeechConfig.FromSubscription(predictionKey, predictionRegion); AudioConfig audioConfig = AudioConfig.FromDefaultMicrophoneInput(); IntentRecognizer recognizer = new IntentRecognizer(speechConfig, audioConfig);

   Python

   Python

   # Configure speech service and get intent recognizer speech_config = speech_sdk.SpeechConfig(subscription=lu_prediction_key, region=lu_prediction_region) audio_config = speech_sdk.AudioConfig(use_default_microphone=True) recognizer = speech_sdk.intent.IntentRecognizer(speech_config, audio_config)

   If you need to use an audio file:

   C#

   C#

   // Configure speech service and get intent recognizer string audioFile = "time-in-london.wav"; SoundPlayer wavPlayer = new SoundPlayer(audioFile); wavPlayer.Play(); System.Threading.Thread.Sleep(2000); SpeechConfig speechConfig = SpeechConfig.FromSubscription(predictionKey, predictionRegion); AudioConfig audioConfig = AudioConfig.FromWavFileInput(audioFile); IntentRecognizer recognizer = new IntentRecognizer(speechConfig, audioConfig);

   Python

   Python

   # Configure speech service and get intent recognizer audioFile = 'time-in-london.wav' playsound(audioFile) speech_config = speech_sdk.SpeechConfig(subscription=lu_prediction_key, region=lu_prediction_region) audio_config = speech_sdk.AudioConfig(filename=audioFile) recognizer = speech_sdk.intent.IntentRecognizer(speech_config, audio_config)

Get a predicted intent from spoken input

Now you're ready to implement code that uses the Speech SDK to get a predicted intent from spoken input.

1. In the Main function, immediately beneath the code you just added, find the comment Get the model from the AppID and add the intents we want to use and add the following code to get your Language Understanding model (based on its App ID) and specify the intents that we want the recognizer to identify.

   C#

   C#

   // Get the model from the AppID and add the intents we want to use var model = LanguageUnderstandingModel.FromAppId(luAppId); recognizer.AddIntent(model, "GetTime", "time"); recognizer.AddIntent(model, "GetDate", "date"); recognizer.AddIntent(model, "GetDay", "day"); recognizer.AddIntent(model, "None", "none");

   Note that you can specify a string-based ID for each intent

   Python

   Python

   # Get the model from the AppID and add the intents we want to use model = speech_sdk.intent.LanguageUnderstandingModel(app_id=lu_app_id) intents = [ (model, "GetTime"), (model, "GetDate"), (model, "GetDay"), (model, "None") ] recognizer.add_intents(intents)

2. Under the comment Process speech input, add the following code, which uses the recognizer to asynchronously call the Language Understanding service with spoken input, and retrieve response. If the response includes a predicted intent, the spoken query, predicted intent, and full JSON response are displayed. Otherwise the code handles the response based on the reason returned.

C#

C

// Process speech input string intent = ""; var result = await recognizer.RecognizeOnceAsync().ConfigureAwait(false); if (result.Reason == ResultReason.RecognizedIntent) { // Intent was identified intent = result.IntentId; Console.WriteLine($"Query: {result.Text}"); Console.WriteLine($"Intent Id: {intent}."); string jsonResponse = result.Properties.GetProperty(PropertyId.LanguageUnderstandingServiceResponse_JsonResult); Console.WriteLine($"JSON Response:\n{jsonResponse}\n"); // Get the first entity (if any) // Apply the appropriate action } else if (result.Reason == ResultReason.RecognizedSpeech) { // Speech was recognized, but no intent was identified. intent = result.Text; Console.Write($"I don't know what {intent} means."); } else if (result.Reason == ResultReason.NoMatch) { // Speech wasn't recognized Console.WriteLine($"Sorry. I didn't understand that."); } else if (result.Reason == ResultReason.Canceled) { // Something went wrong var cancellation = CancellationDetails.FromResult(result); Console.WriteLine($"CANCELED: Reason={cancellation.Reason}"); if (cancellation.Reason == CancellationReason.Error) { Console.WriteLine($"CANCELED: ErrorCode={cancellation.ErrorCode}"); Console.WriteLine($"CANCELED: ErrorDetails={cancellation.ErrorDetails}"); } }

Python

Python

# Process speech input intent = '' result = recognizer.recognize_once_async().get() if result.reason == speech_sdk.ResultReason.RecognizedIntent: intent = result.intent_id print("Query: {}".format(result.text)) print("Intent: {}".format(intent)) json_response = json.loads(result.intent_json) print("JSON Response:\n{}\n".format(json.dumps(json_response, indent=2))) # Get the first entity (if any) # Apply the appropriate action elif result.reason == speech_sdk.ResultReason.RecognizedSpeech: # Speech was recognized, but no intent was identified. intent = result.text print("I don't know what {} means.".format(intent)) elif result.reason == speech_sdk.ResultReason.NoMatch: # Speech wasn't recognized print("Sorry. I didn't understand that.") elif result.reason == speech_sdk.ResultReason.Canceled: # Something went wrong print("Intent recognition canceled: {}".format(result.cancellation_details.reason)) if result.cancellation_details.reason == speech_sdk.CancellationReason.Error: print("Error details: {}".format(result.cancellation_details.error_details))

The code you've added so far identifies the intent, but some intents can reference entities, so you must add code to extract the entity information from the JSON returned by the service.

1. In the code you just added, find the comment Get the first entity (if any) and add the following code beneath it:

C#

C

// Get the first entity (if any) JObject jsonResults = JObject.Parse(jsonResponse); string entityType = ""; string entityValue = ""; if (jsonResults["entities"].HasValues) { JArray entities = new JArray(jsonResults["entities"][0]); entityType = entities[0]["type"].ToString(); entityValue = entities[0]["entity"].ToString(); Console.WriteLine(entityType + ": " + entityValue); }

Python

Python

# Get the first entity (if any) entity_type = '' entity_value = '' if len(json_response["entities"]) > 0: entity_type = json_response["entities"][0]["type"] entity_value = json_response["entities"][0]["entity"] print(entity_type + ': ' + entity_value)

Your code now uses the Language Understanding app to predict an intent as well as any entities that were detected in the input utterance. Your client application must now use that prediction to determine and perform the appropriate action.

1. Beneath the code you just added, find the comment Apply the appropriate action, and add the following code, which checks for intents supported by the application (GetTime, GetDate, and GetDay) and determines if any relevant entities have been detected, before calling an existing function to produce an appropriate response.

C#

C#

// Apply the appropriate action switch (intent) { case "time": var location = "local"; // Check for entities if (entityType == "Location") { location = entityValue; } // Get the time for the specified location var getTimeTask = Task.Run(() => GetTime(location)); string timeResponse = await getTimeTask; Console.WriteLine(timeResponse); break; case "day": var date = DateTime.Today.ToShortDateString(); // Check for entities if (entityType == "Date") { date = entityValue; } // Get the day for the specified date var getDayTask = Task.Run(() => GetDay(date)); string dayResponse = await getDayTask; Console.WriteLine(dayResponse); break; case "date": var day = DateTime.Today.DayOfWeek.ToString(); // Check for entities if (entityType == "Weekday") { day = entityValue; } var getDateTask = Task.Run(() => GetDate(day)); string dateResponse = await getDateTask; Console.WriteLine(dateResponse); break; default: // Some other intent (for example, "None") was predicted Console.WriteLine("You said " + result.Text.ToLower()); if (result.Text.ToLower().Replace(".", "") == "stop") { intent = result.Text; } else { Console.WriteLine("Try asking me for the time, the day, or the date."); } break; }

Python

Python

# Apply the appropriate action if intent == 'GetTime': location = 'local' # Check for entities if entity_type == 'Location': location = entity_value # Get the time for the specified location print(GetTime(location)) elif intent == 'GetDay': date_string = date.today().strftime("%m/%d/%Y") # Check for entities if entity_type == 'Date': date_string = entity_value # Get the day for the specified date print(GetDay(date_string)) elif intent == 'GetDate': day = 'today' # Check for entities if entity_type == 'Weekday': # List entities are lists day = entity_value # Get the date for the specified day print(GetDate(day)) else: # Some other intent (for example, "None") was predicted print('You said {}'.format(result.text)) if result.text.lower().replace('.', '') == 'stop': intent = result.text else: print('Try asking me for the time, the day, or the date.')

Run the client application

1. Save your changes and return to the integrated terminal for the speaking-clock-client folder, and enter the following command to run the program:

   C#

   dotnet run

   Python

   python speaking-clock-client.py

2. If using a microphone, speak utterances aloud to test the application. For example, try the following (re-running the program each time):

   What's the time?

   What time is it?

   What day is it?

   What is the time in London?

   What's the date?

   What date is Sunday?

Note: The logic in the application is deliberately simple, and has a number of limitations, but should serve the purpose of testing the ability for the Language Understanding model to predict intents from spoken input using the Speech SDK. You may have trouble recognizing the GetDay intent with a specific date entity due to the difficulty in verbalizing a date in MM/DD/YYYY format!

More information

To learn more about Speech and Language Understanding integration, see the Speech documentation.