Multi-device Arbitration for Seamless ASR

 

Introduction

The MASA (Multi-device Arbitration for Seamless ASR) service selects and operates a device that is suitable for a user's intention when there are multiple devices. In addition, the Seamless ASR function is applied so that it can operate even when you speak without distinguishing between wake up word and command.

Trigger words (also called keywords, hotwords, or wakeup words) are widely used as a method of running speech recognition (Automatic Speech Recognition, ASR). Recently, trigger words have been developed and applied to products to go beyond the function of running voice recognition simply and enable a user to perform various functions, which can improve his or her voice recognition experience.

 

 

The MASA service provided by LG AI Platform has the following features:

MWA 엔진의 특징

Features	Description
Seamless ASR	Previously, if a user spoke a trigger word, it understood that the trigger word was recognized through the display or alarm sound, and then spoke a command. However, Seamless ASR can recognize the trigger word and command separately If you speak a trigger word and a command in succession.
Selective Trigger-word Alarm(STA)	Depending on the existing voice recognition method and Seamless ASR, the alarm sound of the device can be played selectively.
Server trigger word rejection	The trigger word recognized by the application is verified once again by the server. If it is determined that a trigger word is misrecognized, the server sends a rejection result for the trigger word to prevent device malfunction.
One device selection	If multiple devices, which use the same trigger word, recognize the trigger word at the same time, the user's intention is identified and one device can be selected for operation.

 

Structure

All features of the MASA service are implemented and run on the server. The MASA service receives PCM data and JSON data as input value and sends the result.

 

Examples of Use

The MASA service is a feature that can be used in a variety of ways in our daily life.

For example, if a user says “Hi LG” in a space where home appliances that recognize the trigger word, “Hi LG,”’ are gathered, all home appliances recognize the trigger word.

After recognizing the trigger word, the MASA service identifies the user's direction and distance, selects a home appliance that can be controlled well, and receives feedback from the product.