English | 日本語

swift-voice-input

A Swift package for voice input on iOS / macOS. Protocol-oriented design allows you to swap in multiple speech recognition backends including Apple Speech.

Features

• Protocol-oriented — Abstracts the recognition engine behind the SpeechRecognizer protocol. Plug in Apple Speech, Whisper, local LLMs, and more
• Real-time streaming — Partial results delivered incrementally via AsyncStream. Use partialText for live display
• Floating preview — An Aqua Voice-style overlay shows recognized text in real time above the input field
• DesignSystem compliant — UI components fully follow the token system of swift-design-system
• Two-target layout — VoiceInput (Core) and VoiceInputUI (SwiftUI) are separated. Depend on Core alone when you don't need UI
• Swift Concurrency — Actor-based audio engine and @Observable state management for safe concurrent access

Installation

Add the dependency to Package.swift:

dependencies: [
    .package(url: "https://github.com/no-problem-dev/swift-voice-input.git", .upToNextMajor(from: "1.0.0")),
]

Add to your target:

// Core only (no UI needed)
.product(name: "VoiceInput", package: "swift-voice-input"),

// Core + SwiftUI components
.product(name: "VoiceInputUI", package: "swift-voice-input"),

Quick Start

Basic Usage

import VoiceInput
import VoiceInputUI

struct MyView: View {
    @State private var session = VoiceInputSession()
    @State private var text = ""

    var body: some View {
        VStack {
            TextField("Type here...", text: $text)

            VoiceInputButton(session: session)
        }
        .voiceInputOverlay(session: session) { transcript in
            text = transcript
        }
    }
}

Inline Transcript Preview

Embed a transcript preview directly in the layout flow using InlineTranscriptView:

import VoiceInput
import VoiceInputUI

struct MyView: View {
    @State private var session = VoiceInputSession()
    @State private var text = ""

    var body: some View {
        VStack(alignment: .leading, spacing: 8) {
            HStack {
                TextField("Type here...", text: $text)
                VoiceInputButton(session: session)
            }
            InlineTranscriptView(session: session) { transcript in
                text = transcript
            }
        }
    }
}

Custom Recognition Engine

Implement an Actor conforming to SpeechRecognizer:

actor WhisperRecognizer: SpeechRecognizer {
    let displayName = "Whisper"
    var isAvailable: Bool { true }

    func requestPermissions() async -> Result<Void, SpeechRecognitionError> {
        // Request microphone permission
        .success(())
    }

    func start(locale: Locale) throws -> AsyncStream<SpeechRecognitionResult> {
        // Start recognition with Whisper model
        AsyncStream { _ in }
    }

    func stop() {
        // Stop recognition
    }
}

// Inject at session creation
@State private var session = VoiceInputSession(
    recognizer: WhisperRecognizer()
)

Session API

let session = VoiceInputSession()

session.toggle()           // Toggle start/stop
session.startListening()   // Start
session.stopListening()    // Stop

session.state              // .idle, .requesting, .listening, .processing, .error
session.partialText        // Real-time partial text
session.transcript         // Finalized text

let text = session.confirm() // Confirm text + reset
session.reset()              // Reset

Architecture

VoiceInput (Core)
├── Protocol/
│   ├── SpeechRecognizer         # Recognition engine abstraction protocol (Actor)
│   ├── SpeechRecognitionResult  # .partial / .final result type
│   └── SpeechRecognitionError   # Error type
├── Engine/
│   ├── AppleSpeechRecognizer    # Default Apple Speech implementation
│   └── PermissionRequester      # Microphone & speech recognition permissions
└── Session/
    └── VoiceInputSession        # @Observable state management

VoiceInputUI (SwiftUI + DesignSystem)
├── Button/
│   └── VoiceInputButton         # Microphone toggle button
├── Inline/
│   └── InlineTranscriptView     # Inline transcript preview
└── Overlay/
    └── TranscriptOverlayModifier # .voiceInputOverlay() modifier

Requirements

License

MIT License — see LICENSE for details.

Links

• Issues