Voice Command Translation Using User Lexical Pattern Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inaccurate translation of voice commands by voice assistants leads to loss of productivity and user trust, particularly in enterprise settings, due to the inability of OEMs to correct erroneous speech recognition outside the intended voice skills.
Innovation Solution
A networked environment utilizing a smart voice service with components like a user pattern engine, error detection engine, comparison engine, and user confirmation application, which leverages historical data to generate and correct voice command translations based on user-specific lexical patterns and grammatical analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If voice commands are translated using standard speech recognition services, then the voice assistant can process commands quickly, but translation accuracy deteriorates leading to erroneous speech recognition
Solution Approach 1:
The system performs preliminary actions by analyzing historical data and generating lexical pattern data before the actual voice command translation occurs. This pre-processing of user-specific language patterns enables more accurate translation without slowing down the actual command execution, as the correction logic operates in parallel or pre-computes the necessary correction rules.
Solution Approach 2:
The system implements feedback mechanisms where historical data from previous interactions is continuously analyzed to refine and update lexical pattern data. This feedback loop allows the system to learn from past translation errors and improve accuracy over time, addressing the precision issue without compromising the speed of individual command processing.
2Measurement precision
If OEMs correct erroneous speech recognition outside voice skills, then translation accuracy improves, but system complexity increases and trust is lost
Solution Approach 1:
The patent introduces an intermediary layer in the form of a correction service that operates independently from the voice skill execution. This intermediary receives the translated command, applies corrections based on lexical patterns, and returns the corrected command to the skill service. This separation of concerns maintains accuracy while managing complexity through modular architecture.
Solution Approach 2:
The system implements self-service by automatically generating and applying corrections based on historical data and lexical patterns without requiring manual intervention or complex configuration. The machine learning models automatically adapt to user-specific language patterns, reducing the need for manual programming of correction rules and simplifying system maintenance.
3Measurement precision
If historical data is used to generate lexical patterns, then translation accuracy improves, but data processing time increases
Solution Approach 1:
The system performs preliminary analysis of historical data to generate lexical pattern data in advance. By pre-processing and storing the corrected patterns, the actual translation process can quickly reference these pre-computed patterns without re-analyzing the entire historical dataset, thus maintaining high accuracy while minimizing real-time processing time.
Solution Approach 2:
The system applies local quality by tailoring lexical patterns specifically to each user's language habits and preferences rather than using a generic correction approach. This user-specific customization is achieved by analyzing individual historical data to create personalized patterns, which then enable fast and accurate translation for that specific user without requiring processing of all historical data during each translation operation.
Data Source
AI summary
The present disclosure relates to translation of voice commands using machine learning. Command text corresponding to a voice command can be received, and at least one error can be identified in the command text. A comparison can be performed between the at least one error and at least one lexical pattern corresponding to a user associated with the voice command. Modified command text can be generated based at least in part on the comparison between the at least one error and the at least one lexical pattern. The modified command text can be determined to fail to comprise an additional error.