Voice Recognition Reject Information Generator
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice recognition systems face challenges in accurately distinguishing between intended and unintended voice inputs, leading to misrecognition issues, particularly when the input voice resembles a target word but does not meet the predetermined likelihood threshold.
Innovation Solution
A voice recognition device and system that utilize a reject information generator to determine and store reject information based on input voices that are misrecognized as target words, using a second voice recognition device to verify and correct the recognition, thereby reducing misrecognition by associating reject words with their acoustic features and preventing incorrect target word identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If voice recognition uses a dictionary with acoustic feature values of words, then voice recognition can be performed, but misrecognition occurs when input voice resembles target words but does not meet the likelihood threshold
Solution Approach 1:
The system implements feedback by using the second voice recognition device to evaluate and correct the recognition results from the first voice recognition device. The reject information generator uses this feedback to identify misrecognized words and generate reject information, which is then fed back into the recognition process to prevent similar misrecognition in the future.
Solution Approach 2:
The reject information generator acts as an intermediary between the first and second voice recognition devices. It processes the recognition results, identifies misrecognition cases, generates reject information, and transmits this to the storage device for future reference, thereby mediating the recognition process to improve accuracy.
2Productivity
If the voice recognizer recognizes input voice as target word based on acoustic feature matching, then recognition speed is maintained, but incorrect target word identification occurs when input voice does not intentionally target a specific word
Solution Approach 1:
The system performs preliminary action by pre-storing reject information in the storage device before actual voice recognition occurs. This reject information, generated from previous misrecognition cases, is ready to be used during recognition to prevent similar errors, allowing the system to maintain speed while improving accuracy through pre-computed rejection criteria.
Solution Approach 2:
The reject information generator serves as an intermediary that bridges the first voice recognition device (which maintains recognition speed) and the second voice recognition device (which verifies accuracy). It processes recognition results and generates reject information that mediates between these two processes, enabling both speed and accuracy to coexist.
3Device complexity
If a single voice recognition device processes all recognition tasks, then device complexity is reduced, but misrecognition cannot be effectively corrected
Solution Approach 1:
The system segments the voice recognition function into two distinct devices: a first voice recognition device for initial recognition and a second voice recognition device for verification and correction. This segmentation allows each device to specialize in specific tasks, with the first device maintaining recognition speed and the second device ensuring accuracy through verification.
Solution Approach 2:
The reject information generator acts as an intermediary component that coordinates between the two voice recognition devices. It processes results from the first device, identifies errors, generates corrective reject information, and transmits this to the second device, thereby mediating the interaction between the segmented recognition systems.
Data Source
AI summary
A voice recognition device includes a storage, a voice recognizer, and a reject information generator. The storage stores reject information for use in specifying a voice. The voice recognizer recognizes an input voice as a target word while defining, as one of conditions, a fact that a degree of likelihood between a voice specified by the reject information and the input voice is lower than a predetermined threshold value. In case where the voice recognizer has recognized the input voice as the target word in a predetermined state, the reject information generator determines the reject information, based on the input voice, and stores the reject information in the storage.