Speech Semantic Understanding via Entity Vocabulary Substring Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semantic understanding technologies rely on pre-stored entity vocabularies in dictionary trees, which are inefficient for accurately determining semantics in speech information due to limitations in handling wrong, extra, or missing words, and lack robustness in matching processes.
Innovation Solution
A method and apparatus that acquire a recognition character string from speech information, match it with entity vocabularies using an entity vocabulary library, and determine a semantic result by calculating character similarity through sliding windows and preset thresholds, ensuring strong fault-tolerance for errors in word recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If character string search is used to find entity vocabularies in dictionary tree, then the matching process is simple, but the accuracy is low when wrong, extra, or missing words occur
Solution Approach 1:
The patent segments the recognition character string into multiple substrings of different lengths (e.g., full string, substrings excluding first/last characters, substrings excluding multiple characters from ends). This segmentation allows the system to search for entity vocabularies at multiple granularity levels, improving accuracy when wrong, extra, or missing words occur by finding matches in substrings even when the full string doesn't match exactly.
Solution Approach 2:
The patent performs entity vocabulary searching not only on the complete recognition character string but also on multiple partial substrings. This excessive action of searching beyond the single complete string allows the system to find entity vocabularies even when the recognition contains errors, by matching against partial correct portions of the input.
2Reliability
If exact matching is required for entity vocabularies, then the reliability is high, but the fault-tolerance for recognition errors is poor
Solution Approach 1:
The patent prepares multiple candidate substrings in advance before performing the entity vocabulary search. By pre-segmenting the recognition character string into multiple potential matching portions, the system cushions against recognition errors - if the complete string doesn't match, the pre-prepared substrings provide fallback options to maintain reliable semantic determination despite faults in the original recognition.
Solution Approach 2:
The patent introduces substrings as intermediary elements between the recognition character string and the entity vocabulary database. These substrings act as mediators that can bridge the gap between imperfect recognition results and the exact matching requirements of the entity vocabulary database, enabling fault-tolerant semantic determination.
3Reliability
If multiple substrings are searched for entity vocabularies, then the fault-tolerance improves, but the search time increases
Solution Approach 1:
The patent performs preliminary segmentation of the recognition character string into multiple substrings before the actual entity vocabulary search. This preliminary action organizes the search space in advance, allowing the system to efficiently traverse predefined substring patterns rather than generating them during search, thereby reducing the time penalty associated with multiple searches.
Solution Approach 2:
The patent implements a dynamic search strategy where the system adaptively selects which substrings to search based on the recognition results and entity vocabulary matching outcomes. The search process dynamically adjusts by attempting matches in order of decreasing substring length or based on likelihood, stopping early when a satisfactory match is found, thus optimizing the balance between robustness and search time.
Data Source
AI summary
A semantic understanding method and apparatus, and a device and a storage medium are provided. The method includes: acquiring a recognition character string that matches speech information; acquiring, from an entity vocabulary library, at least one entity vocabulary respectively corresponding to each recognition character in the recognition character string; and according to a situation of each entity vocabulary hitting the recognition character string, determining a matching entity vocabulary as a semantic understanding result of the speech information. By means of the method, insofar as a completely matching entity vocabulary is not acquired, a matching entity vocabulary can still be determined according to an entity vocabulary library, and semantic information of speech is thus accurately understood; and the method also has relatively high fault tolerance for situations such as wrong words, added words, and omitted words, such that the semantic understanding accuracy of speech information is improved.