SwiftSpeechRecognizer

This package contains some very straightforward wrappers around around Speech Recognition part of SFSpeechRecognizer to allow you using this with ease.

SwiftSpeechRecognizer Using Swift Concurrency with async await, a couple of AsyncStream
SwiftSpeechRecognizerDependency A wrapper around the library above facilitating the integration with Point-Free Dependencies library or a project made with The Composable Architecture (TCA).
SwiftSpeechCombine the OG library still available in this package

Authorization

⚠️ For Speech Recognition to be allowed to work, you must add NSSpeechRecognitionUsageDescription in your app Info.plist with a description of its usage.

Modern concurrency usage

import Speech
import SwiftSpeechRecognizer

struct SpeechRecognizer {

    var authorizationStatusChanged: (SFSpeechRecognizerAuthorizationStatus) -> Void
    var speechRecognitionStatusChanged: (SpeechRecognitionStatus) -> Void
    var utteranceChanged: (String) -> Void

    private let speechRecognizer = SwiftSpeechRecognizer.live

    func start() {
        speechRecognizer.requestAuthorization()
        Task {
            for await authorizationStatus in speechRecognizer.authorizationStatus() {
                authorizationStatusChanged(authorizationStatus)

                switch authorizationStatus {
                case .authorized:
                    startRecording()
                default:
                    print("Not authorized to use speech recognizer")
                }
            }
        }
    }

    private func startRecording() {
        do {
            try speechRecognizer.startRecording()
            Task {
                for await recognitionStatus in speechRecognizer.recognitionStatus() {
                    speechRecognitionStatusChanged(recognitionStatus)
                }
            }

            Task {
                for await newUtterance in speechRecognizer.newUtterance() {
                    utteranceChanged(newUtterance)
                }
            }
        } catch {
            print("Something went wrong: \(error)")
            speechRecognizer.stopRecording()
        }
    }
}

import SwiftUI

struct ContentView: View {
    var body: some View {
        Button {
            let speechRecognizer = SpeechRecognizer(
                authorizationStatusChanged: { newAuthorizationStatus in
                    print("Authorization status changed: \(newAuthorizationStatus)")
                },
                speechRecognitionStatusChanged: { newRecognitionStatus in
                    print("Recognition status changed: \(newRecognitionStatus)")
                },
                utteranceChanged: { newUtterance in
                    print("Recognized utterance changed: \(newUtterance)")
                }
            )
            speechRecognizer.start()
        } label: {
            Text("Start speech recognizer")
        }
    }
}

Point-Free Dependency usage

Add @Dependency(\.speechRecognizer) var speechRecognizer in your Reducer, you will have access to all functions mentioned above.

Example

import ComposableArchitecture
import Speech
import SwiftSpeechRecognizerDependency

public struct SpeechRecognizer: ReducerProtocol {
    public struct State: Equatable {
        public var status: SpeechRecognitionStatus = .notStarted
        public var authorizationStatus: SFSpeechRecognizerAuthorizationStatus = .notDetermined
        public var utterance: String?

        public init(
            status: SpeechRecognitionStatus = .notStarted,
            authorizationStatus: SFSpeechRecognizerAuthorizationStatus = .notDetermined,
            utterance: String? = nil
        ) {
            self.status = status
            self.authorizationStatus = authorizationStatus
            self.utterance = utterance
        }
    }

    public enum Action: Equatable {
        case buttonTapped
        case startRecording
        case stopRecording
        case setStatus(SpeechRecognitionStatus)
        case setUtterance(String?)
        case requestAuthorization
        case setAuthorizationStatus(SFSpeechRecognizerAuthorizationStatus)
        case cancel
    }

    @Dependency(\.speechRecognizer) var speechRecognizer

    public init() {}

    private enum AuthorizationStatusTaskID {}
    private enum RecognitionStatusTaskID {}
    private enum NewUtteranceTaskID {}

    public func reduce(into state: inout State, action: Action) -> EffectTask<Action> {
        switch action {
        case .buttonTapped:
            switch state.status {
            case .recording, .stopping:
                return stopRecording(state: &state)
            case .notStarted, .stopped:
                return startRecording(state: &state)
            }
        case .startRecording:
            return startRecording(state: &state)
        case .stopRecording:
            return stopRecording(state: &state)
        case let .setStatus(status):
            state.status = status
            switch status {
            case .recording, .stopping:
                state.utterance = nil
            case .notStarted, .stopped:
                break
            }
            return .none
        case let .setUtterance(utterance):
            state.utterance = utterance
            return .none
        case let .setAuthorizationStatus(authorizationStatus):
            state.authorizationStatus = authorizationStatus
            switch authorizationStatus {
            case .authorized:
                return startRecording(state: &state)
            case .notDetermined:
                return requestAuthorization(state: &state)
            default: break
            }
            return .none
        case .requestAuthorization:
            return requestAuthorization(state: &state)
        case .cancel:
            return .cancel(ids: [AuthorizationStatusTaskID.self, RecognitionStatusTaskID.self, NewUtteranceTaskID.self])
        }
    }

    private func startRecording(state: inout State) -> EffectTask<Action> {
        guard state.authorizationStatus == .authorized
        else { return requestAuthorization(state: &state) }
        do {
            try speechRecognizer.startRecording()
            return .merge(
                .run(operation: { send in
                    for await recognitionStatus in speechRecognizer.recognitionStatus() {
                        await send(.setStatus(recognitionStatus))
                    }
                })
                .cancellable(id: RecognitionStatusTaskID.self),
                .run(operation: { send in
                    for await newUtterance in speechRecognizer.newUtterance() {
                        await send(.setUtterance(newUtterance))
                    }
                })
                .cancellable(id: NewUtteranceTaskID.self)
            )
        } catch {
            return stopRecording(state: &state)
        }
    }

    private func stopRecording(state: inout State) -> EffectTask<Action> {
        speechRecognizer.stopRecording()
        return .none
    }

    private func requestAuthorization(state: inout State) -> EffectTask<Action> {
        speechRecognizer.requestAuthorization()
        return .run { send in
            for await authorizationStatus in speechRecognizer.authorizationStatus() {
                await send(.setAuthorizationStatus(authorizationStatus))
            }
        }
        .cancellable(id: AuthorizationStatusTaskID.self)
    }
}

Combine usage

You can instantiate/inject SpeechRecognitionEngine object, it has this behavior

func requestAuthorization(): call this method when you want to ask user if you can use Speech Recognition. Please, follow Apple Human Guidelines recommendations: Asking Permission to Use Speech Recognition
- subscribe to authorizationStatusPublisher to know when and what's the user decision (see below)
func startRecording() throws: call this method when you want to start the speech recognition, all the complexity with AVAudioEngine is hidden behind this function.
- subscribe to recognizedUtterancePublisher or newUtterancePublisher depending on your needs. See below the difference
func stopRecording(): call this method when you want the speech recognition to stop
var authorizationStatusPublisher: subscribe to this to know what's the current authorization status of the speech recognition (this authorization include the microphone usage)
var recognizedUtterancePublisher: subscribe to this to know when a String is recognized
var isRecognitionAvailablePublisher: subscribe to this to know when the recognition status changed. It could be:
- notStarted: no recognition has started yet for the session
- recording: the recognition is processing
- stopping: a stop recognition has just been triggered
- stopped: the recognition is stopped, and is ready to start the next recognition
var isRecognitionAvailablePublisher: subscribe to this to know when the Speech Recognition is available on the device (like if the internet connection allows the Speech Recognition to be processed)
var newUtterancePublisher: subscribe to this if you're only interested in "new" actual utterance. It's actually a shortcut of recognizedUtterancePublisher while it removes duplicates and only take an actual String (so it's never nil)

Example

import Combine
import Speech
import SwiftSpeechCombine

let engine: SpeechRecognitionEngine = SpeechRecognitionSpeechEngine()
var cancellables = Set<AnyCancellable>()

// Only start to record if you're authorized to do so!
func speechRecognitionStatusChanged(authorizationStatus: SFSpeechRecognizerAuthorizationStatus) {
    guard authorizationStatus == .authorized
    else { return }

    do {
        try engine.startRecording()

        engine.newUtterancePublisher
            .sink { newUtterance in
                // Do whatever you want with the recognized utterance...
                print(newUtterance)
            }
            .store(in: &cancellables)

        engine.recognitionStatusPublisher
            .sink { status in
                // Very useful to notify the user of the current status
                setTheButtonState(status: status)
            }
            .store(in: &cancellables)
    } catch {
        print(error)
        engine.stopRecording()
    }
}

func setTheButtonState(status: SpeechRecognitionStatus) {
    switch status {
    case .notStarted: print("The button is ready to be tapped")
    case .recording: print("The button is showing a progress spinner")
    case .stopping: print("The button is disabled")
    case .stopped: print("The button is ready to be tapped for a new recognition")
    }
}

engine.requestAuthorization()

engine.authorizationStatusPublisher
    .compactMap { $0 }
    .sink { authorizationStatus in
        speechRecognitionStatusChanged(authorizationStatus: authorizationStatus)
    }
    .store(in: &cancellables)

Installation

Xcode

You can add SwiftSpeechCombine to an Xcode project by adding it as a package dependency.

From the File menu, select Swift Packages › Add Package Dependency...
Enter "https://github.com/renaudjenny/swift-speech-recognizer" into the package repository URL test field
Select one of the three libary that you are interested in. See above

As package dependency

Edit your Package.swift to add this library.

let package = Package(
    ...
    dependencies: [
        .package(url: "https://github.com/renaudjenny/SwiftSpeechCombine", from: "1.0.0"),
        ...
    ],
    targets: [
        .target(
            name: "<Your project name>",
            dependencies: [
                .product(name: "SwiftSpeechRecognizer", package: "swift-speech-recognizer"), // <-- Modern concurrency
                .product(name: "SwiftSpeechRecognizerDependency", package: "swift-speech-recognizer"), // <-- Point-Free Dependencies library wrapper
                .product(name: "SwiftSpeechCombine", package: "swift-speech-recognizer"), // <-- Combine wrapper
            ]),
        ...
    ]
)

App using this library

📲 Tell Time UK: https://github.com/renaudjenny/telltime