Voice Processing Combining Similar Sentences for NLU

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing natural language understanding systems face challenges in accurately processing multiple sentences from user voice signals without increasing computational complexity, especially when sentences have insufficient information or repeat similar intents.

Innovation Solution

An electronic apparatus and control method that utilize a processor to receive user voice signals, convert them into text, identify sentences, determine their domains, and combine similar sentences using neural network models to improve natural language understanding while reducing computational load.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a neural network model is used to learn multiple sentences from user voice signals, then the accuracy of natural language understanding is improved, but the model capacity increases and processing speed decreases

Engineering Contradiction:
Improveaccuracy of natural language understandingVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent segments the user voice signal into multiple individual sentences before processing. Each sentence is then independently analyzed for domain and intent, allowing the system to handle complex multi-sentence inputs without requiring a single large neural network model, thus maintaining processing speed while improving understanding accuracy

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary classification of each sentence by domain before conducting detailed natural language understanding. This preliminary action organizes the input sentences into domain-specific groups, enabling more efficient processing by specialized models and improving overall accuracy without significantly increasing computational load

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If a neural network model is used to learn multiple sentences, then the accuracy of identifying user intent is improved, but the computation amount increases

Engineering Contradiction:
Improveaccuracy of identifying user intentVSAvoidcomputation amount
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent divides the computation task by segmenting sentences and processing them individually with domain-specific models. This segmentation allows the system to use smaller, more efficient neural networks for each domain rather than one large comprehensive model, reducing overall computation amount while maintaining intent identification accuracy

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the processing parameters by treating each sentence independently with domain-specific configurations rather than using a single uniform model. This parameter change enables optimized computation for each domain, reducing total computational requirements while improving intent detection accuracy through domain-specialized processing

Inventive Principle:
Principle #35Parameter changes

3Device complexity

If multiple sentences are processed independently, then the processing complexity is reduced, but the accuracy of reflecting user intent deteriorates

Engineering Contradiction:
Improveprocessing complexityVSAvoidaccuracy of reflecting user intent
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary domain classification for each sentence before processing, creating an organized structure that maintains low processing complexity. This preliminary organization enables the system to capture contextual relationships between sentences within the same domain while keeping the overall processing framework simple and manageable

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4254401B1Electronic device and control method therefor
Publication Date: 2025.02.26 SAMSUNG ELECTRONICS CO LTD
  • EP4254401B1 patent drawingFigure 1
  • EP4254401B1 patent drawingFigure 2
  • EP4254401B1 patent drawingFigure 3

AI summary

An electronic device is disclosed. The electronic device comprises a mic; a memory for storing at least one instruction; and a processor connected to the mic and the memory so as to control the electronic device. The processor, by executing the at least one instruction, may: when a user speech signal is received through the mic, acquire text corresponding to the user speech signal; identify a plurality of sentences included in the acquired text; identify, among a plurality of domains, repective domains corresponding to each of the plurality of sentences; if the degree of similarity between a first sentence and a second sentence having the same domain, among the plurality of sentences, is equal to or higher than a threshold value, acquire, using a first neural network model, a third sentence formed by combining the first sentence and the second sentence; and carry out natural language understanding of the third sentence.