Speech Recognition Error Correction via Probability Adjustment

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Spoken dialog systems face challenges in accurately interpreting user intent due to disfluencies and slips of the tongue, which are not effectively addressed by current automatic speech recognition and understanding technologies.

Innovation Solution

The method modifies the probabilities of closely related words in speech recognition models using a corpora of data to identify and correct common misstatements, particularly by adjusting probabilities based on phonetic similarities and domain-specific error patterns, and incorporates a learning mode to adapt to individual speech patterns.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If traditional automatic speech recognition uses statistical models to recognize words, then recognition speed is improved, but accuracy deteriorates when users speak non-fluently with disfluencies and slips of the tongue

Engineering Contradiction:

Improverecognition speedVSAvoidword recognition accuracy

Core Design Contradiction:

SpeedVSMeasurement precision

Solution Approach 1:

The system performs preliminary action by pre-identifying likely speech errors using a corpus of common misstatements before final word recognition. When a speech error is detected, the system proactively adjusts probability distributions to favor correct words, preventing misrecognition before it affects understanding.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements feedback by continuously monitoring speech patterns against the corpus of common errors and dynamically adjusting recognition probabilities in real-time. This feedback loop allows the system to correct itself when disfluencies occur, maintaining high accuracy without sacrificing speed.

Inventive Principle:

Principle #23Feedback

2Adaptability or versatility

If the system accepts a larger set of target words to handle language variations, then adaptability is improved, but recognition accuracy deteriorates

Engineering Contradiction:

Improvelanguage variation handlingVSAvoidword recognition accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system applies local quality by treating different words and contexts differently. Instead of uniform treatment, it identifies specific local patterns (common speech errors) and applies targeted corrections only where needed, while maintaining standard recognition for clear speech. This allows handling of language variations without compromising overall accuracy.

Inventive Principle:

Principle #3Local quality

3Measurement precision

If the system corrects speech errors by modifying word probabilities, then recognition accuracy is improved, but processing time increases

Engineering Contradiction:

Improveword recognition accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system applies partial action by correcting only the specific words identified as likely errors rather than reprocessing the entire speech transcript. This selective correction approach maintains accuracy improvement while minimizing the time penalty, as only localized probability adjustments are made rather than full reprocessing.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS7505906B2System and method for augmenting spoken language understanding by correcting common errors in linguistic performance

Publication Date: 2009.03.17 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

A method and system for automatic speech recognition are disclosed. The method comprises receiving speech from a user, the speech including at least one speech error, increasing the probabilities of closely related words to the at least one speech error and processing the received speech using the increased probabilities. A corpora of data having common words that are mis-stated is used to identify and increase the probabilities of related words. The method applies to at least the automatic speech recognition module and the spoken language understanding module.