Text Correction Model Using Phonetic Features for Varying-Length Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current non-autoregressive models for correcting Chinese speech recognition results struggle with varying-length corrections, leading to relatively low accuracy in correction results.
Innovation Solution
A method that involves acquiring a phonetic symbol sequence for the text to be corrected, and using this sequence in conjunction with a text correction model to detect error words, determine corresponding phonetic symbols, and add phonetic features behind the error words to generate a corrected text.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a non-autoregressive model is used for text correction, then the correction speed is fast, but the accuracy of correction results is low due to inability to process varying-length corrections
Solution Approach 1:
The patent segments the correction process into two distinct phases: an error detection phase using a non-autoregressive model for rapid identification of error positions, and an error correction phase that processes varying-length corrections with autoregressive modeling. This segmentation allows the system to leverage the speed of non-autoregressive processing while achieving accurate varying-length corrections through the subsequent refinement stage.
Solution Approach 2:
The patent performs preliminary error detection using the non-autoregressive model before conducting the actual correction. By first identifying error positions and segments in advance, the system prepares the groundwork for accurate correction while maintaining fast processing speed. This preliminary action separates the speed-critical detection phase from the accuracy-critical correction phase.
2Productivity
If a non-autoregressive model is used for text correction, then the processing efficiency is high, but the ability to handle varying-length corrections is poor
Solution Approach 1:
The patent divides the correction task into error detection (fixed-length processing) and error correction (varying-length processing) segments. The non-autoregressive model handles the fixed-length error detection efficiently, while the autoregressive model handles the varying-length correction generation, thus combining high efficiency with adaptability to varying lengths.
Solution Approach 2:
The patent introduces dynamic length handling by using autoregressive modeling specifically for the correction phase where varying lengths occur. The system dynamically switches between fixed-length detection and variable-length correction, allowing it to adapt to different correction scenarios while maintaining overall processing efficiency through the structured two-phase approach.
Data Source
AI summary
Disclosed are a method for correcting a text, an electronic device and a storage medium. The method includes: acquiring a text to be corrected; acquiring a phonetic symbol sequence of the text to be corrected; and obtaining a corrected text by inputting the text to be corrected and the phonetic symbol sequence into a text correction model, in which, the text correction model obtains the corrected text by: detecting an error word in the text to be corrected, determining a phonetic symbol corresponding to the error word in the phonetic symbol sequence, and adding the phonetic feature corresponding to the phonetic symbol behind the error word to obtain a phonetic symbol text, and correcting the error word and the phonetic feature in the phonetic symbol text to obtain the corrected text.