Hume AI Swift SDK

Integrate Hume APIs directly into your Swift application

Documentation

API reference documentation is available here.

Installation

Adding to your Xcode project

1. Open project settings > Package Dependencies
2. Click the + button to add a package dependency
3. Enter SDK URL (https://github.com/HumeAI/hume-swift-sdk.git)
4. Set the version rule (we recommend pinning to a specific version)
5. Click "Add Package"
6. Add the Privacy - Microphone Usage Description entry to your Info.plist
7. (Optional) If you plan to support background audio, select the "Audio, Airplay, and Picture and Picture" option in the "Background Modes" section of your project capabilities.

Adding to your Package.swift

    dependencies: [
        .package(url: "https://github.com/HumeAI/hume-swift-sdk.git", from: "x.x.x")
    ]

Usage

Voice Chat

The SDK provides a VoiceProvider abstraction that manages active socket connection against the /chat endpoint. This abstraction handles and coordinates the audio stack.

Capabilities

• Pipes output audio from audio_output events into SoundPlayer to play back in realtime.
• VoiceProvider.connect(...) opens and connects to the /chatsocket, waits for the chat_metadata event to be received, and starts the microphone.
• VoiceProvider.disconnect() closes the socket, stops the microphone, and stops all playback.

Example

import Hume

let token = try await myAccessTokenClient.fetchAccessToken()
humeClient = HumeClient(options: .accessToken(token: token))

let voiceProvider = VoiceProviderFactory.getVoiceProvider(client: humeClient)
voiceProvider.delegate = myDelegate

// Request permission to record audio. Be sure to add `Privacy - Microphone Usage Description`
// to your Info.plist

if MicrophonePermission.current == .undetermined {
    let granted = await MicrophonePermission.requestPermissions()
    guard granted else {
        print("user declined mic permsissions")
        return 
    }
} else if MicrophonePermission.current == .denied {
    print("user previously declined mic permissions") // ask user to update in settings
    return 
}

let sessionSettings = SessionSettings(
    systemPrompt: "my optional system prompt",
    variables: ["myCustomVariable": myValue, "datetime": Date().formattedForSessionSettings()])

try await voiceProvider.connect(
    configId: myConfigId,
    configVersion: nil,
    sessionSettings: sessionSettings)

// Sending user text input manually
await self.voiceProvider.sendUserInput(message: "Hey, how are you?")

Listening for VoiceProvider updates

Implement VoiceProviderDelegate methods to be notified of events, errors, meter data, state, etc.

TTS

Example

import Hume

let token = try await myAccessTokenClient.fetchAccessToken()
humeClient = HumeClient(options: .accessToken(token: token))

let ttsClient = humeClient.tts

let postedUtterances: [PostedUtterance] = [PostedUtterance(
    description: voiceDescription,
    speed: speed,
    trailingSilence: trailingSilence,
    text: text,
    voice: .postedUtteranceVoiceWithId(PostedUtteranceVoiceWithId(id: "<config ID>", provider: .humeAi))
)]
let fmt = .wav(FormatWav()
let request = PostedTts(
    context: nil,
    numGenerations: 1,
    splitUtterances: nil,
    stripHeaders: nil,
    utterances: postedUtterances,
    instantMode: true,
    format: fmt)

let stream = tts.synthesizeFileStreaming(request: request)
for try await data in stream {
    // convert data to SoundClip
    guard let soundClip = SoundClip.from(data) else {
        print("warn: failed to create sound clip")
        return
    }
            
    // play SoundClip with ttsPlayer
    try await ttsPlayer.play(soundClip: soundClip, format: fmt)
    _data.append(data)
    
}

Beta Status

This SDK is in beta, and there may be breaking changes between versions without a major version update. Therefore, we recommend pinning the package version to a specific version. This way, you can install the same version each time without breaking changes.

Known Issues and Limitations

• Audio interruptions (e.g. phone calls) are not yet handled.
• Manually starting/stopping AVAudioSession will likely break an active voice session. Leave all audio handling to AudioHub. If you need to add your own output audio nodes, see `AudioHub.addNode(_:)
• Input metering is not yet implemented.