Domain-Specific Text Modification for Speech Recognition Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current speech recognition systems face challenges in accurately modifying speech recognition results across different domains, leading to inconsistencies and inaccuracies in text output, particularly when dealing with diverse voice inputs and contexts.

Innovation Solution

A system and method where a device and server interoperate to utilize an automatic speech recognition (ASR) model and an artificial intelligence (AI) model for text modification, identifying the relevant domain and selecting appropriate text modification models to refine the output text, ensuring accuracy and context relevance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single speech recognition model is used across different domains, then the system complexity is reduced, but the recognition accuracy and context relevance deteriorate in domain-specific scenarios

Engineering Contradiction:
Improvesystem complexityVSAvoidrecognition accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent divides the speech recognition system into multiple domain-specific models (e.g., general domain model, news domain model, sports domain model) instead of using a single unified model. Each model is trained on domain-specific data to improve recognition accuracy for that particular domain while maintaining manageable system complexity through modular architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system dynamically selects which domain-specific model to use based on the input context and domain identification. The model selection is not static but adapts in real-time according to the speech content, allowing the system to optimize between using a simple general model or a specialized domain model as needed.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If domain-specific text modification models are selected and applied, then the text output accuracy and context relevance are improved, but the device complexity and model selection overhead increase

Engineering Contradiction:
Improvetext output accuracyVSAvoiddevice complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs domain identification and model selection in advance before the main speech recognition processing. By pre-classifying the input speech into domains and pre-selecting the appropriate model, the system avoids complex real-time adjustments during recognition, thereby improving text output accuracy while managing complexity through upfront preparation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces a domain identification module as an intermediary between the speech input and the recognition models. This mediator analyzes the input speech to determine the appropriate domain, then routes the input to the corresponding domain-specific model, simplifying the overall system architecture while maintaining high accuracy through specialized processing.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If multiple domain-specific models are maintained and switched between, then the adaptability to different speech contexts is improved, but the loss of time for model switching and selection occurs

Engineering Contradiction:
Improveadaptability to different speech contextsVSAvoidtime for model switching
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs domain identification and model selection in advance before the main speech recognition processing. By pre-classifying the input speech into domains and pre-selecting the appropriate model, the system avoids complex real-time adjustments during recognition, thereby improving text output accuracy while managing complexity through upfront preparation.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3790002B1System and method for modifying speech recognition result
Publication Date: 2024.10.02 SAMSUNG ELECTRONICS CO LTD
  • EP3790002B1 patent drawingFigure 1
  • EP3790002B1 patent drawingFigure 2
  • EP3790002B1 patent drawingFigure 3

AI summary

Provided are a system and method for modifying a speech recognition result. The method includes: receiving, from a device, text output from an automatic speech recognition (ASR) model of the device; identifying at least one domain related to the received text; selecting, from among a plurality of text modification models included in the server, at least one text modification model corresponding to the identified at least one domain; and modifying the received text by using the selected at least one text modification model.