Grammar Correction for Speech Recognition Using Reliability Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems often generate text with grammatical errors due to recognition inaccuracies, and existing grammar correction techniques fail to accurately correct these errors as they do not account for the reliability of the speech recognition process.
Innovation Solution
A grammar correcting method and apparatus that receives a sentence generated by speech recognition, determines recognition reliabilities based on pronunciation and intention, detects candidate words with low reliability, and corrects grammatical errors using a language and acoustic model to generate a more accurate sentence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general techniques for correcting grammatical errors are used, then grammar correction can be performed, but the correction accuracy is insufficient because they do not consider speech recognition reliability
Solution Approach 1:
The correction system is segmented into multiple functional modules: a reliability determination module that assesses speech recognition confidence for each word, a grammar verification module that identifies grammatical errors, and a correction module that applies corrections based on both grammar rules and reliability scores. This segmentation allows the system to process complex correction tasks through manageable stages, improving accuracy while maintaining reasonable complexity
Solution Approach 2:
The system performs preliminary determination of speech recognition reliability for each word in the sentence before conducting grammar correction. By pre-assessing which words have low recognition confidence, the system can prioritize correction efforts on those specific words, thereby improving overall correction accuracy without needing to overhaul the entire correction process
2Productivity
If speech recognition is performed without considering reliability, then text can be generated quickly, but grammatical errors occur due to recognition inaccuracies
Solution Approach 1:
The system applies partial action by determining speech recognition reliability only for words that are suspected to contain errors or have ambiguous recognition. Rather than performing exhaustive reliability analysis on every word in the sentence, the system focuses computational resources on critical words, thereby maintaining text generation speed while improving recognition accuracy for problematic areas
Solution Approach 2:
The system implements feedback by using the determined speech recognition reliability information to guide the grammar correction process. Words with low reliability scores receive higher priority for correction, and the correction module uses this feedback to adjust its correction strategy, thereby improving overall recognition accuracy while maintaining efficient processing
Data Source
AI summary
A grammar correcting method is provided, the method including receiving a sentence generated based on speech recognition, receiving information associated with a speech recognition result of the sentence and correcting grammar in the sentence based on the information associated with the speech recognition results of the sentence.