Sliding Window Speech Correction Module
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General-purpose speech recognition engines often produce inaccurate outputs when recognizing spoken sentences in specific domains due to environmental factors and accent variations, making it necessary to correct the output before natural language processing can be effectively performed.
Innovation Solution
A processor-implemented method that identifies the current environment of a speaker and selectively corrects the output of a general-purpose automatic speech recognition engine by replacing erroneous terms with best-fit terms using a fitness function and pre-determined threshold, through multi-stage correction processes involving domain ontology, contextual, and linguistic corrections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a general-purpose speech recognition engine is used to recognize spoken sentences in specific domains, then the device complexity is reduced and ease of operation is improved, but the accuracy of the recognition output deteriorates due to environmental factors and accent variations
Solution Approach 1:
The patent introduces an intermediary correction module that sits between the general-purpose ASR engine and the natural language processing system. This module receives the ASR output, identifies potential errors using a sliding window mechanism, and corrects them by selecting the best-fit terms from domain-specific vocabularies. The intermediary handles the accuracy issue without requiring a complete domain-specific ASR system, thus maintaining low device complexity while improving measurement precision.
Solution Approach 2:
The system dynamically adjusts parameters such as the sliding window size, confidence thresholds, and fitness function weights based on the identified current environment of the speaker. By changing these parameters adaptively, the system optimizes correction accuracy for different environmental conditions and accents while still using the same general-purpose ASR engine, resolving the contradiction between simplicity and accuracy.
2Measurement precision
If domain-specific ASR engines are used to improve accuracy in specific domains, then the accuracy of speech recognition output is improved, but the device complexity and cost increase
Solution Approach 1:
The patent creates a universal correction framework that can work with any general-purpose ASR engine output. The correction module uses domain-specific vocabularies and environment-based adaptation to handle multiple domains and conditions without requiring separate domain-specific ASR engines. This multi-functional approach achieves domain-specific accuracy while maintaining a single, simpler system architecture.
Solution Approach 2:
The system employs self-service mechanisms where the correction module automatically identifies and corrects errors in ASR output without human intervention. By using automated fitness functions, sliding window analysis, and environment-based adaptation, the system serves its own correction needs, eliminating the need for complex manual domain-specific ASR configurations and reducing overall device complexity.
3Measurement precision
If post-processing correction methods are applied to GP-ASR output, then the accuracy is improved, but the processing time and loss of time increase
Solution Approach 1:
The sliding window correction mechanism applies partial action by focusing corrections only on specific portions of the ASR output where errors are most likely to occur. Rather than reprocessing the entire output, the system identifies error-prone regions and applies corrections selectively, reducing the overall processing time while maintaining high accuracy for the corrected portions.
Solution Approach 2:
The system performs preliminary actions by pre-processing domain-specific vocabularies and organizing them for quick access during correction. The fitness functions and environment variables are pre-configured based on the identified current environment, allowing the correction process to proceed efficiently without extensive real-time computation, thus minimizing time loss while improving accuracy.
Data Source
AI summary
Text output of speech recognition engines tend to be erroneous when spoken data has domain specific terms. The present disclosure facilitates automatic correction of errors in speech to text conversion using abstractions of evolutionary development and artificial development. The words in a speech recognition engine text output are treated as a set of injured genes in a biological cell that need repair which are then repaired and form genotypes that are then repaired to phenotypes through a series of repair steps based on a matching, mapping and linguistic repair through a fitness criteria. A basic genetic level repair involves phonetic MATCHING function together with a FITNESS function to select the best among the matching genes. A second genetic level repair involves a contextual MAPPING function for repairing remaining ‘injured’ genes of the speech recognition engine output. Finally, a genotype to phenotype repair involves using linguistic rules and semantic rules of the domain.