Speech Recognition Accuracy via Context Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech recognition technologies face challenges in accurately transcribing speech to text, particularly when words have low reliability, leading to poor recognition rates due to inappropriate text correction in the context of the entire text.
Innovation Solution
A speech recognition method that performs multiple speech recognition processes for sections with low reliability, verifies similarities between recognition candidates and meta-information, and determines the recognition result based on these similarities, using techniques like doc2vec for context verification and deep learning to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional predictive transform is used to correct text with low reliability words, then text correction is performed, but the correction is not appropriate in terms of context or text as a whole leading to poor recognition rate
Solution Approach 1:
The system performs multiple speech recognition processes and uses the results to verify similarities between recognition candidates and meta-information. This feedback mechanism allows the system to evaluate and select the most appropriate recognition result based on contextual consistency, thereby improving speech recognition accuracy while maintaining reliable recognition rates
Solution Approach 2:
Instead of applying a single correction method, the system performs multiple speech recognition processes and selects from multiple recognition candidates. This partial application of different recognition methods allows the system to choose the most appropriate result for each context, improving overall recognition accuracy without compromising reliability
2Measurement precision
If multiple speech recognition processes are performed for low-reliability sections, then recognition accuracy is improved, but processing complexity increases
Solution Approach 1:
The system segments the speech recognition process by identifying low-reliability sections and applying multiple recognition processes specifically to those sections. This segmentation allows the system to improve recognition accuracy for problematic areas without unnecessarily increasing processing complexity for the entire speech input
Solution Approach 2:
The system applies different processing quality to different sections of the speech input. Low-reliability sections receive multiple recognition processes and thorough verification, while high-reliability sections are processed more simply. This local quality approach improves overall accuracy while controlling processing complexity
Data Source
AI summary
A speech recognition method, performed by a computer, with an improved recognition accuracy is disclosed. The method includes: performing speech recognition of an input speech to acquire a plurality of recognition candidates through a plurality of speech recognition processes different from each other for a section having a reliability lower than a predetermined value; verifying similarities between each of the acquired plurality of recognition candidates and meta-information corresponding to the input speech; and determining, based on the verified similarities, a recognition result of the low-reliability section from among the acquired plurality of recognition candidates.