Dynamic Keyword Model Assignment for Voice Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional mobile devices have limited sound models and keywords, leading to inaccurate detection of new keywords due to insufficient sampling, restricting user functionality and convenience.

Innovation Solution

A method and apparatus for assigning a target keyword to a function in electronic devices, where a list of keywords is received via a communication network, and a keyword model is downloaded and used to detect the target keyword, enabling accurate activation of functions upon detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If users generate sound models by training based on limited utterances, then new keywords can be detected, but detection accuracy is low due to insufficient sampling

Engineering Contradiction:
Improvekeyword detection capabilityVSAvoidkeyword detection accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by pre-training sound models using a large number of utterances collected in advance from multiple users. Instead of training with limited user utterances at the time of keyword addition, the system prepares comprehensive sound models beforehand, ensuring high detection accuracy when new keywords are introduced to the device.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If manufacturers equip devices with pre-configured sound models, then keyword detection is reliable, but the number of detectable keywords is limited

Engineering Contradiction:
Improvekeyword detection accuracyVSAvoidnumber of detectable keywords
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamics by enabling the sound model to dynamically adapt and expand its keyword recognition capability. The system maintains a base sound model with pre-trained keywords and allows dynamic addition of new keywords through user input. The sound model evolves from a static, limited set to a dynamic, expandable vocabulary while preserving detection accuracy through the pre-training foundation.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent applies universality by creating a sound model that serves multiple functions: it recognizes both pre-configured keywords and user-defined keywords. The single sound model structure handles diverse keyword types and functions, making the system versatile without requiring separate models for different keyword categories.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If extensive sampling is used to train accurate sound models, then detection accuracy improves, but power consumption and training time increase

Engineering Contradiction:
Improvekeyword detection accuracyVSAvoidpower consumption during training
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent resolves this contradiction by performing the extensive sampling and model training in advance, during device manufacturing or initial setup. The computationally intensive work of processing large numbers of utterances is completed beforehand, allowing the device to operate with a ready-to-use, accurate sound model without consuming excessive power during normal operation.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3020040B1Method and apparatus for assigning keyword model to voice operated function
Publication Date: 2018.12.19 QUALCOMM INC
  • EP3020040B1 patent drawingFigure 1
  • EP3020040B1 patent drawingFigure 2
  • EP3020040B1 patent drawingFigure 3

AI summary

A method, performed in an electronic device, for assigning a target keyword to a function is disclosed. In this method, a list of a plurality of target keywords is received at the electronic device via a communication network, and a particular target keyword is selected from the list of target keywords. Further, the method may include receiving a keyword model for the particular target keyword via the communication network. In this method, the particular target keyword is assigned to a function of the electronic device such that the function is performed in response to detecting the particular target keyword based on the keyword model in an input sound received at the electronic device.