Post-Processing Speech System for ASR Error Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current automatic speech recognition (ASR) technologies face significant error rates and security vulnerabilities, particularly when handling critical information or urgent conditions, and struggle to recognize spoken utterances at the sentence level, leading to potential interception of sensitive data.

Innovation Solution

A post-processing speech system that employs a grammar-based ASR architecture with real-time processing and normalization, using a parallel tree-based structure to support both limited and unlimited vocabulary sizes, and integrates domain-specific grammars to enhance accuracy and security, while maintaining language agnosticism and scalability across audio and multi-modal systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If natural language-based ASR is used to recognize spoken utterances, then vocabulary coverage is improved, but error rate increases and security vulnerabilities arise

Engineering Contradiction:
Improvevocabulary coverageVSAvoiderror rate
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The system segments the ASR process into two distinct passes: a first pass using natural language processing for broad vocabulary coverage, and a second pass using grammar-based verification for accuracy and security. This segmentation allows each pass to specialize - the first handles unknown vocabulary while the second ensures precise recognition and prevents errors, thereby resolving the contradiction between vocabulary coverage and error rate

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary verification mechanism where grammar-based rules act as a mediator between the natural language ASR output and the final recognition result. This intermediary layer validates and corrects the initial recognition, filtering out errors while preserving the broad vocabulary capability of the natural language approach

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If natural language-based ASR is used, then language flexibility is improved, but security vulnerabilities and interception risks increase

Engineering Contradiction:
Improvelanguage flexibilityVSAvoidsecurity vulnerabilities
Core Design Contradiction:
Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

The system performs preliminary grammar-based verification before finalizing the ASR output. By applying security rules and validation constraints in advance (in the second pass), the system prevents security vulnerabilities and interception risks before they can affect the final result, while still allowing language flexibility in the initial recognition phase

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If grammar-based ASR is used, then accuracy is improved, but vocabulary coverage is limited

Engineering Contradiction:
Improverecognition accuracyVSAvoidvocabulary coverage
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent merges two previously separate ASR approaches into a unified two-pass system. The first pass combines natural language processing capabilities for broad vocabulary coverage, while the second pass integrates grammar-based verification for high accuracy. The merging allows the system to achieve both broad vocabulary coverage and high recognition accuracy simultaneously

Inventive Principle:
Principle #5Merging (Combining)

4Adaptability or versatility

If continuous training is implemented, then system adaptability is improved, but processing resources increase substantially

Engineering Contradiction:
Improvesystem adaptabilityVSAvoidprocessing resources
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

Instead of implementing continuous training across the entire system, the patent applies partial action by using grammar-based verification only for critical security and accuracy requirements. This selective approach provides sufficient adaptability for the specific application while avoiding the substantial processing resources required for comprehensive continuous training

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2660810B1Post processing of natural language ASR
Publication Date: 2019.10.02 ONTARIO INC
  • EP2660810B1 patent drawingFigure 1
  • EP2660810B1 patent drawingFigure 2
  • EP2660810B1 patent drawingFigure 3

AI summary

A post-processing speech system includes a natural language-based speech recognition system that compares a spoken utterance to a natural language vocabulary that includes words used to generate a natural language speech recognition result. A master conversation module engine compares the natural language speech recognition result to domain specific words and phrases. A voting engine selects a word or a phrase from the domain specific words and phrases that is transmitted to an application control system. The application control system transmits one or more control signals that are used to control an internal or an external device or an internal or an external process.