Punctuation Mark Delete Model for Speech Recognition Text
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Text obtained from speech recognition often contains errors, leading to incorrect assignment of punctuation marks, and existing technologies lack accuracy in punctuation assignment due to the absence of voice information in the processing.
Innovation Solution
A punctuation mark delete model learning device is developed to generate a punctuation mark delete model through machine learning, which determines the correctness of punctuation marks in speech recognition text by using pairs of sentences with labeled correct punctuation assignments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If punctuation marks are assigned based only on text and pattern matching, then the processing is simple and fast, but the accuracy of punctuation mark assignment is low
Solution Approach 1:
The patent combines text-based pattern matching with voice information analysis to determine punctuation mark placement. The punctuation mark assignment unit integrates both the arrangement pattern of phrases and dependency relationships between phrases, along with voice characteristics, to achieve more accurate punctuation assignment than text-only methods.
Solution Approach 2:
The patent introduces a punctuation mark delete model as an intermediary component that processes the output from speech recognition. This model uses learned patterns from training data to identify and correct incorrect punctuation marks, acting as a mediator between raw speech recognition output and final corrected text.
2Measurement precision
If punctuation marks are assigned at the end of sentence division based on text only, then the processing is straightforward, but the accuracy is low because voice information is not referred to
Solution Approach 1:
The patent performs preliminary action by training a punctuation mark delete model in advance using training data that includes both text and voice information. The model learns the relationship between voice characteristics and correct punctuation placement before being applied to actual speech recognition results, preserving voice information effects in the punctuation assignment process.
Solution Approach 2:
The patent segments the punctuation assignment process into multiple stages: initial punctuation assignment based on text patterns, followed by correction using the punctuation mark delete model that incorporates voice information. This segmentation allows systematic integration of voice data without completely redesigning the punctuation assignment system.
3Reliability
If speech recognition text is used directly without correction, then the processing is fast and simple, but punctuation marks are sometimes assigned to incorrect places due to recognition errors
Solution Approach 1:
The patent applies partial correction by using the punctuation mark delete model to identify and correct only the incorrect punctuation marks in speech recognition output, rather than completely reassigning all punctuation marks. This selective approach maintains processing efficiency while improving correctness.
Solution Approach 2:
The patent implements feedback by using the punctuation mark delete model to evaluate and correct punctuation marks in speech recognition results. The model provides feedback on which punctuation marks are incorrect based on learned patterns from training data, enabling iterative improvement of punctuation accuracy.
Data Source
AI summary
A punctuation mark delete model learning device is a device that generates, through machine learning, a punctuation mark delete model, and comprises a first learning data generation unit that generates first learning data consisting of a pair of an input sentence including a punctuation mark, a preceding sentence that is a sentence with the punctuation mark assigned at an end of the sentence, and a subsequent sentence following the punctuation mark, and a label indicating whether or not the assignment of the punctuation mark is correct, on the basis of a first text corpus consisting of text obtained by speech recognition processing, and a model learning unit that updates parameters of the punctuation mark delete model on the basis of an error between a probability obtained by inputting the input sentences of the first learning data to the punctuation mark delete model and the label.