Speech Recognition Error Correction via Difference Thresholding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Speech recognition systems often produce errors due to environmental factors and unknown words, which existing correction technologies fail to address effectively, leading to inefficient user corrections and high correction costs.

Innovation Solution

A speech recognition error correction apparatus that utilizes a correction network memory and error correction circuitry to associate speech recognition results with user corrections, calculating differences and performing error corrections only when the difference is below a threshold, thereby reducing repetitive corrections and costs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If speech recognition is performed in various environments, then recognition coverage is improved, but recognition accuracy deteriorates due to errors from environmental factors and unknown words

Engineering Contradiction:
Improverecognition coverageVSAvoidrecognition accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent introduces a correction network as an intermediary component between the speech recognition system and the final output. This correction network, built from historical correction data, mediates the recognition results by automatically correcting errors without requiring re-recognition, thus maintaining recognition coverage while improving accuracy through error correction

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system performs preliminary actions by collecting and storing correction data from historical speech recognition tasks. This preliminary data collection and correction network construction enables the system to proactively prepare correction patterns before new recognition tasks, allowing automatic correction of recurring errors in various environments

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If manual correction is performed for each speech recognition error, then recognition accuracy is improved, but correction time and cost increase

Engineering Contradiction:
Improverecognition accuracyVSAvoidcorrection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The correction network enables the speech recognition system to self-correct errors automatically. By utilizing the correction network that stores historical correction patterns, the system performs self-service error correction without requiring manual intervention for each error, thus improving accuracy while reducing correction time

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system copies successful correction patterns from historical data into the correction network. Instead of manually correcting each new error, the system replicates proven correction approaches from the correction network, enabling automatic correction of similar errors and significantly reducing repetitive manual correction time

Inventive Principle:
Principle #26Copying

Data Source

PatentUS11620981B2Speech recognition error correction apparatus
Publication Date: 2023.04.04 KK TOSHIBA
  • US11620981B2 patent drawing
  • US11620981B2 patent drawing
  • US11620981B2 patent drawing

AI summary

According to one embodiment, a speech recognition error correction apparatus includes a correction network memory and an error correction circuitry. The error correction circuitry calculates a difference between a speech recognition result string of an error correction target, which is a result of performing speech recognition on a new series of speech data, and a correction network, where a speech recognition result string and a correction result by a user for the speech recognition result string are associated, and when a value indicating the difference is equal to or less than a threshold, perform error correction on a speech recognition error portion in the speech recognition result string of the error correction target by using the correction network to generate a speech recognition error correction result string.