Getting Started

Edit on GitHub

This guide will get you up and running with Spokestack for iOS, and you’ll be hearing and responding to your users in no time. We’ll start with how to integrate speech recognition, and then move on to how to process that recognized speech using natural langauge understanding, and finally learn how to speak back to your users using text to speech.

One caveat before we start, though: This is not a collection of best practices. We’re going to be trading thoughtful organization for convenience here, so when we say something like “put this in your view controller”, just know that you might not want to leave it there long-term. OK, now that that’s out of the way, let’s jump right in.


Your app needs to target iOS 11 or higher in order to use speech recognition and natural language understanding, and iOS 13 or higher in order to use text to speech.


The rest of the guide will assume you already have the Spokestack framework installed in your project. See the README for more information, but if you’re using CocoaPods, it’s as easy as adding

pod 'Spokestack-iOS'
platform :ios, '13.0'

to your Podfile and running

pod install

in your terminal.


In order for your app to accept voice input via Spokestack, it needs four things:

  1. The proper iOS permissions
  2. An active AVAudioSession
  3. An instance of Spokestack’s SpeechPipeline
  4. Delegates to receive system events from SpeechPipeline

1. Permissions

For the first one, head over to Info.plist in your project and add a couple keys. Here are the raw values you’ll need (the keys, at least; feel free to substitute your own values):

<string>The microphone is used to receive user commands via voice.</string>
<string>Speech recognition is used to translate user voice input into text for further processing.</string>

If you’re not using “Raw Keys & Values” mode in Xcode, the keys are Privacy - Microphone Usage Description and Privacy - Speech Recognition Usage Description, respectively. Having these description strings means that your app will prompt the user for the appropriate permissions the first time you activate the speech pipeline. Setting up that pipeline is our next step.

2. AVAudioSession

Apple manages the various demands on a phone’s audio system via audio sessions. See their documentation for more details, but here’s the minimum configuration you’ll need in order to record user speech. A good place for it is likely your AppDelegate’s application(_:didFinishLaunchingWithOptions) method.

import AVFoundation


func application(_ application: UIApplication, didFinishLaunchingWithOptions launchOptions:
    [UIApplication.LaunchOptionsKey: Any]?) -> Bool {
    let audioSession = AVAudioSession.sharedInstance()
    do {
        try audioSession.setCategory(AVAudioSession.Category.playAndRecord, mode: AVAudioSession.Mode.default,
                                     options: [AVAudioSession.CategoryOptions.defaultToSpeaker, AVAudioSession.CategoryOptions.allowBluetooth, AVAudioSession.CategoryOptions.allowAirPlay])
        try audioSession.setActive(true)
    } catch let error as NSError {
        // handle error

3. Speech Pipeline and Controllers

With the proper permissions in place, it’s time to decide where you’d like to receive and process speech input and output. In a single-view app, the easiest place for this is going to be your main view controller. import Spokestack at the top of the file, and add SpeechPipeline, TensorflowNLU, and TextToSpeech class members:

public let pipeline = SpeechPipelineBuilder()
public let tts = TextToSpeech(self, configuration: pipeline.configuration)
public let nlu = try! NLUTensorflow(self, configuration: pipeline.configuration)

Note that these components must persist outside the scope of the calling function, so don’t declare it inside a function call that will get garbage collected! If this is confusing, please consult the fuller discussion of the pipeline. Then, after things are loaded:


There are many options for configuring the speech pipeline. This particular setup will begin capturing audio when pipeline.start() is called and use a Voice Activity Detection (VAD) component to send any audio determined to be speech through on-device ASR using Apple’s SFSpeech API. In other words, the app is always actively listening, and no wakeword detection is performed. See the configuration guide for more information about pipeline building options. Using a vadTriggerAppleSpeech profile is a good way to test out ASR without having to tap a button to activate it or downloading and configuring wakeword models. Consider your use-case fully before using it in production, however, since it will capture all speech it hears, not just what’s directed at your app.

The self in this example means that the class containing this pipeline also adopts PipelineDelegate, SpeechEventListener, NLUDelegate, and TextToSpeechDelegate which, conveniently enough, are the next steps.

4. Implementing Delegates

Now that we have a pipeline for speech and controllers for NLU and TTS, we need to provide implementations for the delegates so that they can send events to you. We’ll do that in the same class we used in the previous step.

If you want to disable any buttons or show a special “listening” indicator while recording, put those things in the PipelineDelegate’s didStart and didStop methods; otherwise, the main methods you’ll want to implement are SpeechEventListener’s’ didActivate, didDeactivate, and didRecognize. The basic layout for the first two would be:

// assumes `self` is the class from before; it
// has a `SpeechPipeline` member
func didActivate() {

func didDeactivate() {

All we’re doing here is reflecting system events back to the main pipeline. See the SpeechPipeline guide for more discussion.

To wake or not to wake

Now that you have the pipeline all set up, how do you use it? It’s easy, but the answer depends on your app’s needs:

I want the user to tap a button before talking

We’ve already configured the speech pipeline to use a tap-to-talk option! After the pipeline is started, call pipeline.activate() in the action of whatever button you want to activate the microphone. This skips the wakeword step of the pipeline and starts the Automatic Speech Recognition (ASR) component directly. ASR will stop automatically after the user is silent for a few seconds (how many seconds is one of the configuration parameters we hinted at earlier) or after a preconfigured timeout is reached, but if you need to stop listening immediately for any reason, call pipeline.deactivate(). You can then call pipeline.activate() to start ASR again or pipeline.stop() to shut the pipeline down completely.

Note that, as we mentioned earlier, the very first time you start a speech pipeline, the microphone is activated, so your user will be presented with permissions modals for the microphone and speech recognition; you may want to plan for this in your designs.

I want to use a wakeword

If you want your app to be controllable purely by voice, you need a wakeword — a word (or short phrase) that tells your app “the next thing the user says is meant for you”. Spokestack comes with a default wakeword (“Spokestack”, believe it or not), and that’s enabled just by changing the pipeline profile enum in the SpeechPipeline we just set up. Try changing .vadTriggerAppleSpeech to .appleWakewordAppleSpeech in that first code example. Then, to begin listening for it, just call pipeline.start().

Understanding your users

Inside SpeechEventListener’s’ didRecognize delegate function, the result.transcript will give you the raw text of what the user just said. Translating that raw text into an action in your app is the job of an NLU, or natural language understanding, component. Spokestack leaves the choice of NLU up to you, but we do offer our own full-featured NLU component for Spokestack based on years of research and lessons learned from working with other services. Our NLU runs directly on your user’s device, instead of calling back to the cloud. If you’re a fan of the cloud, though, you might want to check out DialogFlow, LUIS, or And if your app is simple enough, you can make your own NLU with string matching or regular expressions (see the cookbook for an example of this).

Let’s run through a quick usage of Spokestack’s TensorflowNLU. In SpeechEventListener’s’ didRecognize, ask the NLU to classify what the ASR has recognized:

class MyViewController: UIViewController, SpeechEventListener, NLUDelegate {

    // ...other SpeechEventListener functions...

    func didRecognize(_ result: SpeechContext) {
        let userText = result.transcript
        let classification = self.nlu.classify(utterance: userText)

Earlier, if you recall, we claimed that MyViewController already implemented the NLUDelegate protocol. The nlu.classify call above will return the classification results to a function in that delegate.

    // ...other delegate functions...

    func classification(result: NLUResult) {
        let intent = result.intent
        let slots = result.slots
        switch result.intent {
        // using the example of a timer
        case "start":
            // start the timer and change the UI accordingly
        case "stop":
            // stop the timer and change the UI accordingly
        case "reset":
            // reset the timer and change the UI accordingly
        case default:
            // for when the model doesn't match the intent

You’ll note that the intents are just a single string. A intent-based classifier will regularize all sorts of related langague into a single canonical intent, eg “let’s go” or “please cease” get classified as start and stop, respectively.

Talking back to your users

If you want full hands-free and eyes-free interaction, you’ll want to deliver responses via voice as well. This requires a text-to-speech (TTS) component, and Spokestack has one of these too!

It’s completely separate from the SpeechPipeline, so you can even use it to talk to a user at any point. First, you’ll need to adopt the TextToSpeechDelegate protocol somewhere—say, your view controller:

import AVFoundation


class MyViewController: UIViewController, SpeechEventListener, TextToSpeechDelegate {

    func didBeginSpeaking() {
        // handle the response playback beginning if desired

    func didFinishSpeaking() {
        // handle the response playback ending if desired

    func success(result: TextToSpeechResult) {
        // handle the result if desired

    func failure(error: Error) {
        // handle error

    func didTrace(_ trace: String) {
        // log trace. Note that tracing verbosity of each component is determined by the SpeechConfiguration.tracing setting!

To play the voice response represented by your input with the default audio system, just call the speak function!

// with your properties
// assumes `self` adopts `TextToSpeechDelegate`
// uses default SpeechConfiguration values for api access.
let tts = TextToSpeech(self, configuration: SpeechConfiguration())


func speak(_ text: String) {
    let input = TextToSpeechInput(text)

The speak function will call your delegate’s didBeginSpeaking and didFinishSpeaking at the start and finish, respectively, of response playback.

In this example, SpeechConfiguration.apiId and SpeechConfiguration.apiSecret are set to sample values that let you try Spokestack TTS with a demo voice, without creating an account. You can get your own free API credentials. For more TTS input configuration options, see the TTS guide.

If you want more fine-grained control over how the TTS response is played back, you’re free to feed the TextToSpeechResult.url in the success handler into your own audio player. See the cookbook for a quick version of that recipe.


That’s all there is to it! Your app is now configured to accept voice commands and play back synthetic voice responses. Obviously there’s more we could tell you, and you can have much more control over the speech recognition process (including, but not limited to, configuring the pipeline’s sensitivity, using a different ASR provider, or adding your own custom wakeword model). If you’re interested in these advanced topics, check out our other guides. We’ll be adding to them as Spokestack grows.

Thanks for reading!