Speech Recognition Reference Pattern Adaptation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional speech recognition reference pattern adaptation methods fail to fully utilize input speech features, particularly those with low reliability factors, and suffer from low adaptation performance when the original recognition accuracy is low, as only high-reliability data influences adaptation, neglecting information from low-reliability data.

Innovation Solution

A method that calculates adaptation data using recognition error knowledge to correct and utilize input speech data, even when recognition accuracy is inferior, by statistically analyzing recognition errors and clustering speech data to extract detailed error knowledge for effective reference pattern adaptation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If only high-reliability speech data is used for adaptation, then adaptation reliability is improved, but information utilization is worsened

Engineering Contradiction:

Improveadaptation reliabilityVSAvoidinformation utilization

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent converts the harmful effect of low-reliability data (which would normally be discarded) into a benefit by using recognition error knowledge to correct and utilize this data for adaptation. The error knowledge extracted from low-reliability data allows the system to learn from it, transforming what was previously wasted information into a useful resource for improving the reference pattern.

Inventive Principle:

Principle #22Blessing in disguise (Convert harm into benefit)

Solution Approach 2:

The patent changes the parameter of data reliability by introducing recognition error knowledge as a correction factor. Instead of simply filtering data based on reliability thresholds, the system adjusts the adaptation process using error knowledge parameters to weight and correct low-reliability data, allowing it to contribute meaningfully to the adaptation while maintaining overall reliability.

Inventive Principle:

Principle #35Parameter changes

2Productivity

If recognition error knowledge is extracted and used for adaptation, then adaptation performance is improved, but device complexity is worsened

Engineering Contradiction:

Improveadaptation performanceVSAvoiddevice complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by extracting recognition error knowledge in advance from training data before the actual adaptation process. This pre-extracted error knowledge is then stored and reused during adaptation, eliminating the need to perform complex error analysis in real-time and reducing the computational burden during the adaptation phase.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces recognition error knowledge as an intermediary element between the raw speech data and the adaptation process. This intermediary captures the error patterns from recognition results and translates them into correction information, simplifying the overall system architecture by separating the error analysis function from the main adaptation algorithm.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Loss of information

If all input speech data is used for adaptation, then information utilization is improved, but adaptation reliability is worsened

Engineering Contradiction:

Improveinformation utilizationVSAvoidadaptation reliability

Core Design Contradiction:

Loss of informationVSReliability

Solution Approach 1:

The patent applies local quality by treating different speech data segments differently based on their reliability characteristics. Through clustering and error knowledge extraction, the system identifies local regions of high and low reliability within the data and applies appropriate weighting and correction factors to each region, allowing low-reliability data to be used with appropriate caution while maintaining overall adaptation reliability.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS8762148B2Reference pattern adaptation apparatus, reference pattern adaptation method and reference pattern adaptation program

Publication Date: 2014.06.24 NEC CORP

AI summary

A method and apparatus for carrying out adaptation using input speech data information even at a low reference pattern recognition performance. A reference pattern adaptation device 2 includes a speech recognition section 18, an adaptation data calculating section 19 and a reference pattern adaptation section 20. The speech recognition section 18 calculates a recognition result teacher label from the input speech data and the reference pattern. The adaptation data calculating section 19 calculates adaptation data composed of a teacher label and speech data. The adaptation data is composed of the input speech data and the recognition result teacher label corrected for adaptation by the recognition error knowledge which is the statistical information of the tendency towards recognition errors of the reference pattern. The reference pattern adaptation section 20 adapts the reference pattern using the adaptation data to generate an adaptation pattern.