Voice Input System Sifting Space Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voice input systems on handset devices face issues such as excessive candidate options leading to increased user time and error rates due to a large searching space and lack of effective methods to reduce recognition errors, especially in noisy environments.
Innovation Solution
A voice input system and method that combines a voice recognition button with a sifting condition button to narrow the searching space using a speech recognition model, incorporating a resifting mechanism to further refine recognition results, thereby reducing insertion, deletion, and substitution errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the voice input system uses a large searching space constructed by acoustic model, lexical model and language model, then the voice recognition can cover more candidate words, but the user will waste time sifting through too many candidate words and have unpleasant experience
Solution Approach 1:
The patent segments the voice recognition process into multiple stages: initial voice recognition to generate candidate words, then separate sifting processes based on different parameters (noise level, candidate count, accuracy requirements). This segmentation allows the system to handle large searching spaces efficiently by breaking down the sifting task into manageable steps rather than presenting all candidates at once.
Solution Approach 2:
The patent implements dynamic adjustment of the searching space based on real-time parameters. The system dynamically modifies the voice recognition process by applying different sifting conditions (noise thresholds, candidate word limits, accuracy requirements) to adapt the searching space size according to specific recognition needs, thereby reducing unnecessary sifting while maintaining recognition coverage.
2Adaptability or versatility
If the voice input system lacks more restriction on the searching space, then the voice recognition can be more flexible, but the recognition rate decreases and insertion error, deletion error and substitution error increase
Solution Approach 1:
The patent applies parameter changes to optimize the searching space by introducing multiple controllable parameters: noise level thresholds, candidate word count limits, and accuracy requirements. These parameters dynamically adjust the restriction level of the searching space, allowing the system to maintain flexibility while improving recognition rate and reducing errors through optimized parameter settings.
3Ease of operation
If the voice input system provides a plurality of candidate words for user sifting, then the user can choose the correct word, but if the candidate words are too many or recognition consequences are listed too many, the user will waste time and have unpleasant experience
Solution Approach 1:
The patent implements feedback mechanisms where the system monitors user interactions with candidate words and adjusts subsequent recognition processes accordingly. By analyzing user selection patterns and sifting behavior, the system provides feedback to optimize the candidate word list generation, reducing the number of candidates presented while ensuring the correct word remains selectable, thereby decreasing user sifting time without compromising choice capability.
Data Source
AI summary
In the present invention, a voice input system and a voice input method are provided. The voice input method includes the steps of: (A) initiating a speech recognition process by a first input associated with a first parameter of a first speech recognition subject; (B) providing a voice and a searching space constructed by a speech recognition model associated with the first speech recognition subject; (C) obtaining a sub-searching space from the searching space based on the first parameter; (D) searching at least one candidate item associated with the voice from the sub-searching space; and (E) showing the at least one candidate item.