Edge Trigger Word Detection



Edge Trigger Word Detection (ETWD) Engine determines the presence of a pre-specified trigger word in voice input by the user.

The ETWD engine is an embedded-based recognition engine that can be used by selecting either TWD (Trigger Word Detection) or NTWD (Neural Trigger Word Detection). The details are as follows.

  • TWD engine
    Supervised learning with re-recorded data from the target product for loading into the product is finally optimized by over-fitting.


  • NTWD engine
    Since it is a structure that learns a DNN-based phoneme unit model independent of the target start word and transfers the target actuation word, the learning time is relatively reduced and it has strong characteristics against product/channel/environment deviation. The NTWD engine uses a neural network-based algorithm, so it requires a lot more computation than the TWD engine, so it is necessary to check whether the target is operating or not.


Image for the keyword detection process of HTWD Engine


ETWD Engine by LG AI Platform supports the following features:

HTWD Engine feature
Feature Description
Support for various trigger words

hilg, heycloi, airstar

Multi-language support

Supports Korean and English languages


Engine Architecture

ETWD Engine combines embedded trigger word detection with server-based trigger word detection.


etwd architecture


Examples of Use

The ETWD function can be used in all devices equipped with a speech recognition function. This engine allows a user to initiate speech recognition without touching the device.


  • Devices with a virtual assistant

You can control the device by the voice from a mobile device equipped with an AI virtual assistant system.

Image for devices with voice assistant system

  • Home appliances capable of speech recognition​

It can be applied to operate home appliances with voice recognition-capability.

Image for the home appliance with speech recognition

  • Robot control

Operate and control robots using natural language commands.

Image for the robot device