Voice Recognition Dictionary Using Positional Proximity Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice recognition devices struggle to accurately recognize input voices that include combinations of words not adjacent to each other in a text, due to limitations in dictionary generation and language model training, resulting in low precision recognition.
Innovation Solution
A recognizing device that generates a dictionary associating words with positional information, allowing for the calculation of a connection score based on the proximity of connected words, enabling precise recognition of non-adjacent word combinations by comparing input voice signals with character strings and determining corresponding strings based on similarity and connection scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a dictionary is generated using all connection patterns of words in a text, then the recognition precision for non-adjacent word combinations is improved, but the amount of information in the dictionary becomes excessively large
Solution Approach 1:
The patent segments the connection patterns into two categories: adjacent word patterns (registered in the dictionary) and non-adjacent word patterns (generated on-demand during recognition). This segmentation allows the dictionary to remain compact while still supporting recognition of non-adjacent word combinations by calculating connection scores dynamically based on positional information.
Solution Approach 2:
The patent performs preliminary action by storing positional information of words in the dictionary during dictionary generation. This pre-stored positional information enables efficient calculation of connection scores for non-adjacent word patterns during recognition without requiring those patterns to be pre-registered in the dictionary, thus resolving the contradiction between recognition precision and dictionary size.
2Measurement precision
If N-gram language model re-training is performed to increase probability of connected words, then the recognition precision is improved, but the computational complexity increases exponentially
Solution Approach 1:
The patent applies partial action by implementing a simplified version of language model re-training that focuses only on calculating connection scores based on positional information, rather than performing full N-gram re-training. This partial implementation achieves improved recognition precision for connected words without incurring the exponential computational complexity of complete N-gram model re-training.
Data Source
AI summary
A recognizing device includes a memory and a processor coupled to the memory. The memory stores words included in a text and positional information indicating a position of the words in the text. The processor executes a process including comparing an input voice signal with reading information of a character string that connects a plurality of words stored in the memory to calculate a similarity; calculating a connection score indicating a proximity between the plurality of connected words based on positional information of the words stored in the memory; and determining a character string corresponding to the voice signal based on the similarity and the connection score.