Live Understanding

Updated: Apr 15, 2026

Overview

The Wit.ai speech endpoint can return a partial Automatic Speech Recognition (ASR) response as it analyzes audio data. Instead of a single ASR response being sent from the speech endpoint after the entire audio transmission is processed, Live Understanding enables you to get an initial response and potentially act upon that response more quickly.

Voice SDK fires OnWitPartialResponse events with initial ASR response data, similar to how transcriptions are provided via OnPartialTranscription and OnFullTranscription.

Voice SDK provides an AppVoiceExperience method called AcceptPartialResponseAndCancelRequest. When called from an OnWitPartialResponse event listener, this method marks the partial response as the final result. AppVoiceExperience then stops listening and uses that result immediately.

This process has also been integrated with intent/entity matchers for greater effectiveness. Once the confidence score with the matcher reaches the preset threshold, the partial response is automatically marked as the final response data.

Live Understanding applies only to the speech endpoint. It does not affect the message endpoint used for text analysis, since that endpoint processes the complete text in a single pass.

The simplest way to integrate Live Understanding into your app is to use matchers, which help match intents and entities to the user utterance, and then accept partial responses once they pass the matcher criteria.

For developers who need custom control over partial responses, Voice SDK also exposes the OnWitPartialResponse event directly. You can add an event listener and call AcceptPartialResponseAndCancelRequest to accept or reject individual partial responses based on your own criteria.

Prerequisites

Voice SDK installed and configured in your Unreal Engine project

An AppVoiceExperience Blueprint in your project

A Wit.ai app configured with intents and entities

For setup instructions, see Integrate Voice SDK.

Integrate Live Understanding

In the Unreal editor, open the AppVoiceExperience Blueprint to edit it.

Click + Add Component and add a VoiceIntentWithEntityMatcher.

On the Details panel, set the Intent and Entity to match the expectation you have for the match. For example, if your app changes the color of items on command, you might set Intent to set_color, and Entity to color:color.

Under Partial Response, enable the Is Also Used for Partial Response option.

Select the Auto Accept Partial Response Once Past Matcher Criteria option.

If needed, change the Intent Confidence Threshold and Entity Confidence Threshold to match your needs. The maximum value is 1.0, minimum value is 0.0.