ASR Alphanumeric Correction via Phonetic Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic Speech Recognition (ASR) engines often produce erroneous outputs, particularly for alphanumeric sequences, which can be critical in applications like e-commerce, where accuracy is vital, and conventional error correction techniques rely on human intervention or predefined vocabularies, failing to address context and domain-specific errors effectively.
Innovation Solution
A computer-implemented method that extracts erroneous alphanumeric sequences from ASR outputs, generates candidates using phonetic similarity, ranks them based on scores, and selects the highest-ranked candidate for correction, leveraging statistical parameters and domain rules to enhance accuracy without relying on original sound or language-specific dictionaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional error correction techniques are used, then human intervention is required, but correction accuracy for domain-specific alphanumeric sequences deteriorates
Solution Approach 1:
The system performs automatic error correction without human intervention by using phonetic similarity scoring and domain-specific rules to identify and correct erroneous alphanumeric sequences autonomously
Solution Approach 2:
The patent replaces manual human correction with an automated computational system that uses phonetic analysis, statistical scoring, and domain rules to perform correction tasks
2Device complexity
If predefined vocabulary sets are used for correction, then correction process is simplified, but ability to handle domain-specific errors deteriorates
Solution Approach 1:
The system changes from using fixed predefined vocabulary sets to using dynamic phonetic similarity scores and domain-specific parameters (such as alphanumeric sequence patterns and contextual rules) to adapt to different correction scenarios
Solution Approach 2:
The correction system transitions from static predefined vocabularies to a dynamic scoring mechanism that evaluates multiple candidate corrections based on phonetic similarity and domain rules, allowing flexible adaptation to various error types
3Productivity
If lattice traversal of known vocabulary words is used, then search space is limited, but correction accuracy for unknown alphanumeric sequences deteriorates
Solution Approach 1:
The patent introduces phonetic similarity as an intermediary mechanism that bridges the gap between known vocabulary and unknown alphanumeric sequences, allowing the system to evaluate candidate corrections based on phonetic matching rather than requiring exact vocabulary matches
Solution Approach 2:
The system adds a new dimension to the search space by incorporating phonetic similarity scoring, transforming the correction problem from a simple vocabulary lookup to a multi-dimensional evaluation considering phonetic distance, domain rules, and contextual relevance
Data Source
AI summary
Provided are techniques for domain specific correction of output from automatic speech recognition. An output of an automatic speech recognition engine is received. An alphanumeric sequence is extracted from the output, where the alphanumeric sequence represents an erroneous translation by the automatic speech recognition engine. Candidates for the alphanumeric sequence are generated. The candidates are ranked based on scores associated with the candidates. A candidate of the candidates having a highest score of the scores is selected. The output is corrected by replacing the alphanumeric sequence with the selected candidate. The corrected output is returned.