Voice Signal Content Matching from Incomplete STT Text
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech-to-text (STT) technologies produce incomplete and inaccurate results, making it difficult to determine relevant content accurately due to typos and incompleteness, particularly in identifying key words or target words.
Innovation Solution
A method utilizing a computing device with a neural network to perform speech-to-text conversion, followed by topic segmentation and keyword extraction, and then comparing the results against user-specific dictionaries and thresholds to identify target words, ultimately recommending associated content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If speech-to-text conversion is performed using existing STT technology, then voice signal can be converted to text, but the result text contains typos and inscriptions reducing accuracy
Solution Approach 1:
The patent introduces an intermediary process between STT conversion and content determination: a keyword extraction module that uses user profile information, topic segmentation, and candidate keyword generation to mediate the inaccurate STT results. This intermediary layer filters and corrects STT output by comparing against user-specific topic models and selecting most relevant keywords, thus resolving the accuracy issue while maintaining conversion efficiency
Solution Approach 2:
The patent replaces the mechanical/straightforward STT text matching approach with an intelligent system using machine learning models (topic segmentation models, keyword extraction models). Instead of directly using STT output for content determination, the system substitutes a multi-stage processing pipeline involving neural networks and probabilistic models to extract keywords, thereby improving accuracy without sacrificing the initial conversion capability
2Productivity
If keyword identification is performed directly on incomplete STT result text, then content determination can proceed, but accuracy is low due to incompleteness
Solution Approach 1:
The patent performs preliminary actions before final content determination: it first segments the STT text into topics using topic segmentation models, then extracts candidate keywords from each topic, and finally selects the most relevant keywords based on user profiles. This preliminary multi-stage processing prepares the data in advance, enabling accurate keyword identification even from incomplete STT results while maintaining efficient content determination
Solution Approach 2:
The patent segments the incomplete STT result text into multiple topic sections using topic segmentation models. By dividing the text into meaningful topics first, the system can then extract keywords from each segment independently, improving overall keyword identification accuracy. This segmentation approach allows the system to work effectively with incomplete STT output by processing manageable, context-rich segments rather than the entire incomplete text at once
3Device complexity
If a simple STT result text approach is used, then the system is simple to implement, but it is difficult to determine whether key words or target words are included
Solution Approach 1:
The patent implements self-service mechanisms where the system automatically builds user-specific topic models and keyword extraction rules based on user profile information without requiring manual intervention. The system serves itself by learning from user data, automatically segmenting topics, extracting keywords, and refining its own performance. This automation reduces implementation complexity while improving keyword detection capability through adaptive, user-specific processing
Solution Approach 2:
The patent changes key parameters of the processing system dynamically: it adjusts topic segmentation parameters, keyword extraction thresholds, and candidate selection criteria based on user profiles and context. By making these parameters adaptive rather than fixed, the system becomes easier to implement (automatic adjustment) while improving keyword detection accuracy (context-appropriate parameters). The parameter changes enable the system to handle different users and contexts without manual reconfiguration
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed is a method for determining a content associated with a voice signal, which is performed by a computing device. The method may include: converting a voice signal and generating text information; determining a plurality of target word candidates; determining a target word among the plurality of target word candidates based on a comparison between the plurality of target word candidates and the generated text information, and determining a content associated with the target word.