Intelligent Media Services:Initiate a call using a wake word

Background

For smart hardware, a wake word is often used to activate a voice assistant for an online conversation. For example:

• The user says the wake word, waits for the device to respond (such as "I'm here"), and then gives a command.

  User: Hey, Siri.
  Siri: I'm here.
  User: What's the weather like today?
  Siri: It's cloudy today...

• The user says the wake word and the question in a single phrase, and the device responds directly to the question.

  User: Hey, Siri, what's the weather like today?
  Siri: It's cloudy today...

This guide is intended for hardware devices that have integrated a wake up SDK. It explains how to initiate a call upon detecting the wake word to ensure an immediate response.

Prerequisite

• You have integrated the call agent and implemented basic voice call features. For more information, see Integration overview.

Implementation

Core process:

• Pre-fetch token: Your application server pre-fetches the required call token and cache it on the device. This reduces network request latency when initiating a call.

• Pre-start audio capture: Enable RTC audio capture and pre-join capture buffering when creating the call engine. This shortens the time to the first upstream audio frame.

• Mutual exclusion: When a call starts, the wake up SDK (especially its microphone capture) must be stopped. Restart the wake word function after the call ends.

• Response handling:

  • For a normal start (playing a response such as "I'm here"):

    • Option 1: Use the RTC engine to play the prompt tone. This leverages its built-in acoustic echo cancellation (AEC) to prevent self-capture.

    • Option 2: Play the prompt tone from the application layer. During playback, you must mute the RTC microphone to prevent the sound from being re-recorded.

  • For a Q&A start: The question captured with the wake word is passed as a parameter to the agent to trigger an immediate answer.

The following sections explain how to call the SDK's APIs to implement these key steps.

// Disable the agent's welcome message. This is not needed if no welcome message is configured in the workflow.
artcaiCallConfig.agentConfig.agentGreeting = "";
// For a Q&A start, set wakeUpQuery to the question asked.
artcaiCallConfig.agentConfig.wakeUpQuery = "What's the weather like today?";

// Set other parameters.
...

// Initialize the configuration.
mAICallEngine.init(artcaiCallConfig);

// Start the call.
mAICallEngine.call(token);

@Override
public void onAliRtcEngineCreated(AliRtcEngine engine) {
    // Cache the engine instance for later use.
    mRtcEngine = engine;

    // Start audio capture in advance so that audio can be captured during the connection process.
    engine.startAudioCapture();
    engine.startAudioPlayer();
    // Enable pre-join capture buffering for the RTC.
    engine.setParameter("{\"audio\":{\"user_specified_ahead_push_stream\":true}}");
}

// Mute the microphone.
mAICallEngine.muteMicrophone(true);

// Unmute the microphone.
mAICallEngine.muteMicrophone(false);