Quickstart: Create a voice-first virtual assistant with the Speech SDK, UWP

Quickstarts are also available for speech-to-text, text-to-speech and speech-translation.

In this article, you'll develop a C# Universal Windows Platform (UWP) application by using the Speech SDK. The program will connect to a previously authored and configured bot to enable a voice-first virtual assistant experience from the client application. The application is built with the Speech SDK NuGet Package and Microsoft Visual Studio 2017 or later (any edition).


The Universal Windows Platform lets you develop apps that run on any device that supports Windows 10, including PCs, Xbox, Surface Hub, and other devices.


This quickstart requires:

Optional: Get started fast

This quickstart will describe, step by step, how to make a simple client application to connect to your speech-enabled bot. If you prefer to dive right in, the complete, ready-to-compile source code used in this quickstart is available in the Speech SDK Samples under the quickstart folder.

Create a Visual Studio project

  1. Start Visual Studio 2019.

  2. Make sure the Universal Windows Platform development workload is available. Choose Tools > Get Tools and Features from the Visual Studio menu bar to open the Visual Studio installer. If this workload is already enabled, close the dialog box.

    Screenshot of Visual Studio installer, with Workloads tab highlighted

    Otherwise, select the box next to .NET cross-platform development, and select Modify at the lower right corner of the dialog box. Installation of the new feature takes a moment.

  3. Create a blank Visual C# Universal Windows app. First, choose File > New > Project from the menu. In the New Project dialog box, expand Installed > Visual C# > Windows Universal in the left pane. Then select Blank App (Universal Windows). For the project name, enter helloworld.

    Screenshot of New Project dialog box

  4. The Speech SDK requires that your application is built for the Windows 10 Fall Creators Update or later. In the New Universal Windows Platform Project window that pops up, choose Windows 10 Fall Creators Update (10.0; Build 16299) as Minimum version. In the Target version box, select this version or any later version, and then click OK.

    Screenshot of the New Universal Windows Platform Project window

  5. If you're running 64-bit Windows, you can switch your build platform to x64 by using the drop-down menu in the Visual Studio toolbar. (64-bit Windows can run 32-bit applications, so you can leave it set to x86 if you prefer.)

    Screenshot of Visual Studio toolbar, with x64 highlighted


    The Speech SDK only supports Intel-compatible processors. ARM is currently not supported.

  6. Install and reference the Speech SDK NuGet package. In Solution Explorer, right-click the solution, and select Manage NuGet Packages for Solution.

    Screenshot of Solution Explorer, with Manage NuGet Packages for Solution option highlighted

  7. In the upper-right corner, in the Package Source field, select nuget.org. Search for the Microsoft.CognitiveServices.Speech package, and install it into the helloworld project.

    Screenshot of Manage Packages for Solution dialog box

  8. Accept the displayed license to begin installation of the NuGet package.

    Screenshot of License Acceptance dialog box

  9. The following output line appears in the Package Manager console.

    Successfully installed 'Microsoft.CognitiveServices.Speech 1.5.0' to helloworld
  10. Because the application uses the microphone for speech input, add the Microphone capability to the project. In Solution Explorer, double-click Package.appxmanifest to edit your application manifest. Then switch to the Capabilities tab, select the box for the Microphone capability, and save your changes.

    Screenshot of Visual Studio application manifest, with Capabilities and Microphone highlighted

Add sample code

  1. The application's user interface is defined by using XAML. Open MainPage.xaml in Solution Explorer. In the designer's XAML view, replace the entire contents with the below.

        Background="{ThemeResource ApplicationPageBackgroundThemeBrush}">
            <StackPanel Orientation="Vertical" HorizontalAlignment="Center"  Margin="20,50,0,0" VerticalAlignment="Center" Width="800">
                <Button x:Name="EnableMicrophoneButton" Content="Enable Microphone"  Margin="0,0,10,0" Click="EnableMicrophone_ButtonClicked" Height="35"/>
                <Button x:Name="ListenButton" Content="Talk to your bot" Margin="0,10,10,0" Click="ListenButton_ButtonClicked" Height="35"/>
                <StackPanel x:Name="StatusPanel" Orientation="Vertical" RelativePanel.AlignBottomWithPanel="True" RelativePanel.AlignRightWithPanel="True" RelativePanel.AlignLeftWithPanel="True">
                    <TextBlock x:Name="StatusLabel" Margin="0,10,10,0" TextWrapping="Wrap" Text="Status:" FontSize="20"/>
                    <Border x:Name="StatusBorder" Margin="0,0,0,0">
                        <ScrollViewer VerticalScrollMode="Auto"  VerticalScrollBarVisibility="Auto" MaxHeight="200">
                            <!-- Use LiveSetting to enable screen readers to announce the status update. -->
                            <TextBlock x:Name="StatusBlock" FontWeight="Bold" AutomationProperties.LiveSetting="Assertive"
                    MaxWidth="{Binding ElementName=Splitter, Path=ActualWidth}" Margin="10,10,10,20" TextWrapping="Wrap"  />
            <MediaElement x:Name="mediaElement"/>
  2. Open the code-behind source file MainPage.xaml.cs. You'll find it grouped under MainPage.xaml. Replace the contents with the code below. Here's what this sample covers:

    • Using statements for the Speech and Speech.Dialog namespaces
    • A simple implementation to ensure microphone access, wired to a button handler
    • Basic UI helpers to present messages and errors in the application
    • A landing point for the initialization code path that will be populated later
    • A helper to play back text-to-speech (without streaming support)
    • An empty button handler to start listening that will be populated later
    using Microsoft.CognitiveServices.Speech;
    using Microsoft.CognitiveServices.Speech.Audio;
    using Microsoft.CognitiveServices.Speech.Dialog;
    using System;
    using System.Diagnostics;
    using System.IO;
    using System.Text;
    using Windows.Foundation;
    using Windows.Storage.Streams;
    using Windows.UI.Xaml;
    using Windows.UI.Xaml.Controls;
    using Windows.UI.Xaml.Media;
    namespace helloworld
        public sealed partial class MainPage : Page
            private DialogServiceConnector connector;
            private enum NotifyType
            public MainPage()
            private async void EnableMicrophone_ButtonClicked(object sender, RoutedEventArgs e)
                bool isMicAvailable = true;
                    var mediaCapture = new Windows.Media.Capture.MediaCapture();
                    var settings = new Windows.Media.Capture.MediaCaptureInitializationSettings();
                    settings.StreamingCaptureMode = Windows.Media.Capture.StreamingCaptureMode.Audio;
                    await mediaCapture.InitializeAsync(settings);
                catch (Exception)
                    isMicAvailable = false;
                if (!isMicAvailable)
                    await Windows.System.Launcher.LaunchUriAsync(new Uri("ms-settings:privacy-microphone"));
                    NotifyUser("Microphone was enabled", NotifyType.StatusMessage);
            private void NotifyUser(string strMessage, NotifyType type = NotifyType.StatusMessage)
                // If called from the UI thread, then update immediately.
                // Otherwise, schedule a task on the UI thread to perform the update.
                if (Dispatcher.HasThreadAccess)
                    UpdateStatus(strMessage, type);
                    var task = Dispatcher.RunAsync(Windows.UI.Core.CoreDispatcherPriority.Normal, () => UpdateStatus(strMessage, type));
            private void UpdateStatus(string strMessage, NotifyType type)
                switch (type)
                    case NotifyType.StatusMessage:
                        StatusBorder.Background = new SolidColorBrush(Windows.UI.Colors.Green);
                    case NotifyType.ErrorMessage:
                        StatusBorder.Background = new SolidColorBrush(Windows.UI.Colors.Red);
                StatusBlock.Text += string.IsNullOrEmpty(StatusBlock.Text) ? strMessage : "\n" + strMessage;
                if (!string.IsNullOrEmpty(StatusBlock.Text))
                    StatusBorder.Visibility = Visibility.Visible;
                    StatusPanel.Visibility = Visibility.Visible;
                    StatusBorder.Visibility = Visibility.Collapsed;
                    StatusPanel.Visibility = Visibility.Collapsed;
                // Raise an event if necessary to enable a screen reader to announce the status update.
                var peer = Windows.UI.Xaml.Automation.Peers.FrameworkElementAutomationPeer.FromElement(StatusBlock);
                if (peer != null)
            // Waits for accumulates all audio associated with a given PullAudioOutputStream and then plays it to the
            // MediaElement. Long spoken audio will create extra latency and a streaming playback solution (that plays
            // audio while it continues to be received) should be used -- see the samples for examples of this.
            private void SynchronouslyPlayActivityAudio(PullAudioOutputStream activityAudio)
                var playbackStreamWithHeader = new MemoryStream();
                playbackStreamWithHeader.Write(Encoding.ASCII.GetBytes("RIFF"), 0, 4); // ChunkID
                playbackStreamWithHeader.Write(BitConverter.GetBytes(UInt32.MaxValue), 0, 4); // ChunkSize: max
                playbackStreamWithHeader.Write(Encoding.ASCII.GetBytes("WAVE"), 0, 4); // Format
                playbackStreamWithHeader.Write(Encoding.ASCII.GetBytes("fmt "), 0, 4); // Subchunk1ID
                playbackStreamWithHeader.Write(BitConverter.GetBytes(16), 0, 4); // Subchunk1Size: PCM
                playbackStreamWithHeader.Write(BitConverter.GetBytes(1), 0, 2); // AudioFormat: PCM
                playbackStreamWithHeader.Write(BitConverter.GetBytes(1), 0, 2); // NumChannels: mono
                playbackStreamWithHeader.Write(BitConverter.GetBytes(16000), 0, 4); // SampleRate: 16kHz
                playbackStreamWithHeader.Write(BitConverter.GetBytes(32000), 0, 4); // ByteRate
                playbackStreamWithHeader.Write(BitConverter.GetBytes(2), 0, 2); // BlockAlign
                playbackStreamWithHeader.Write(BitConverter.GetBytes(16), 0, 2); // BitsPerSample: 16-bit
                playbackStreamWithHeader.Write(Encoding.ASCII.GetBytes("data"), 0, 4); // Subchunk2ID
                playbackStreamWithHeader.Write(BitConverter.GetBytes(UInt32.MaxValue), 0, 4); // Subchunk2Size
                byte[] pullBuffer = new byte[2056];
                uint lastRead = 0;
                    lastRead = activityAudio.Read(pullBuffer);
                    playbackStreamWithHeader.Write(pullBuffer, 0, (int)lastRead);
                while (lastRead == pullBuffer.Length);
                var task = Dispatcher.RunAsync(Windows.UI.Core.CoreDispatcherPriority.Normal, () =>
                    mediaElement.SetSource(playbackStreamWithHeader.AsRandomAccessStream(), "audio/wav");
            private void InitializeDialogServiceConnector()
                // New code will go here
            private async void ListenButton_ButtonClicked(object sender, RoutedEventArgs e)
                // New code will go here
  3. Next, you'll create the DialogServiceConnector with your subscription information. Add the following to the method body of InitializeDialogServiceConnector, replacing the strings YourChannelSecret, YourSpeechSubscriptionKey, and YourServiceRegion with your own values for your bot, speech subscription, and region.


    Direct Line Speech (Preview) is currently available in a subset of Speech Services regions. Please refer to the list of supported regions for voice-first virtual assistants and ensure your resources are deployed in one of those regions.


    For information on configuring your bot and retrieving a channel secret, see the Bot Framework documentation for the Direct Line Speech channel.

    // create a DialogServiceConfig by providing a bot secret key and Cognitive Services subscription key
    // the RecoLanguage property is optional (default en-US); note that only en-US is supported in Preview
    const string channelSecret = "YourChannelSecret"; // Your channel secret
    const string speechSubscriptionKey = "YourSpeechSubscriptionKey"; // Your subscription key
    const string region = "YourServiceRegion"; // Your subscription service region. Note: only a subset of regions are currently supported
    var botConfig = DialogServiceConfig.FromBotSecret(channelSecret, speechSubscriptionKey, region);
    botConfig.SetProperty(PropertyId.SpeechServiceConnection_RecoLanguage, "en-US");
    connector = new DialogServiceConnector(botConfig);
  4. DialogServiceConnector relies on several events to communicate its bot activities, speech recognition results, and other information. Add handlers for these events, appending the following to the end of the method body of InitializeDialogServiceConnector.

    // ActivityReceived is the main way your bot will communicate with the client and uses bot framework activities
    connector.ActivityReceived += async (sender, activityReceivedEventArgs) =>
        NotifyUser($"Activity received, hasAudio={activityReceivedEventArgs.HasAudio} activity={activityReceivedEventArgs.Activity}");
        if (activityReceivedEventArgs.HasAudio)
    // Canceled will be signaled when a turn is aborted or experiences an error condition
    connector.Canceled += (sender, canceledEventArgs) =>
        NotifyUser($"Canceled, reason={canceledEventArgs.Reason}");
        if (canceledEventArgs.Reason == CancellationReason.Error)
            NotifyUser($"Error: code={canceledEventArgs.ErrorCode}, details={canceledEventArgs.ErrorDetails}");
    // Recognizing (not 'Recognized') will provide the intermediate recognized text while an audio stream is being processed
    connector.Recognizing += (sender, recognitionEventArgs) =>
        NotifyUser($"Recognizing! in-progress text={recognitionEventArgs.Result.Text}");
    // Recognized (not 'Recognizing') will provide the final recognized text once audio capture is completed
    connector.Recognized += (sender, recognitionEventArgs) =>
        NotifyUser($"Final speech-to-text result: '{recognitionEventArgs.Result.Text}'");
    // SessionStarted will notify when audio begins flowing to the service for a turn
    connector.SessionStarted += (sender, sessionEventArgs) =>
        NotifyUser($"Now Listening! Session started, id={sessionEventArgs.SessionId}");
    // SessionStopped will notify when a turn is complete and it's safe to begin listening again
    connector.SessionStopped += (sender, sessionEventArgs) =>
        NotifyUser($"Listening complete. Session ended, id={sessionEventArgs.SessionId}");
  5. With the configuration established and the event handlers registered, the DialogServiceConnector now just needs to listen. Add the following to the body of the ListenButton_ButtonClicked method in the MainPage class.

    private async void ListenButton_ButtonClicked(object sender, RoutedEventArgs e)
        if (connector == null)
            // Optional step to speed up first interaction: if not called, connection happens automatically on first use
            var connectTask = connector.ConnectAsync();
            // Start sending audio to your speech-enabled bot
            var listenTask = connector.ListenOnceAsync();
            // You can also send activities to your bot as JSON strings -- Microsoft.Bot.Schema can simplify this
            string speakActivity = @"{""type"":""message"",""text"":""Greeting Message"", ""speak"":""Hello there!""}";
            await connector.SendActivityAsync(speakActivity);
        catch (Exception ex)
            NotifyUser($"Exception: {ex.ToString()}", NotifyType.ErrorMessage);
  6. Save all changes to the project.

Build and run the app

  1. Build the application. From the menu bar of Visual Studio, select Build > Build Solution. The code should compile without errors now.

    Screenshot of Visual Studio application, with Build Solution option highlighted

  2. Start the application. From the menu bar of Visual Studio, select Debug > Start Debugging, or press F5.

    Screenshot of Visual Studio application, with Start Debugging option highlighted

  3. A window pops up. In your application, select Enable Microphone, and acknowledge the permission request that pops up.

    Screenshot of permission request

  4. Select Talk to your bot, and speak an English phrase or sentence into your device's microphone. Your speech is transmitted to the Direct Line Speech channel and transcribed to text, which appears in the window.

    Screenshot of successful bot turn

Next steps

See also