Spoken Command Privacy Filtering for Sensitive Data Masking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech recognition equipped smart devices inadvertently transmit and/or record sensitive personal information (SPI) during generic conversations or accidental keyword activations, leading to privacy risks and potential identity theft.
Innovation Solution
A method and system using an AI model to identify and tag SPI in user speech, requiring user confirmation or adherence to predetermined rules before transmitting modified commands to third-party services, masking unnecessary information and anonymizing sensitive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If speech recognition devices proactively listen for buzzwords and transmit user speech to external servers, then the device can respond to user commands efficiently, but sensitive personal information may be inadvertently transmitted and recorded
Solution Approach 1:
The system performs preliminary classification of user speech into generic conversation or command before transmission. It also pre-identifies potential SPI in unmasked portions of classified commands and presents masking options to users before the speech is transmitted to third-party services, preventing inadvertent disclosure
Solution Approach 2:
The patent introduces an intermediary processing layer between the speech recognition device and third-party services. This intermediary classifies speech, identifies potential SPI, presents masking options to users, and only transmits sanitized commands, thereby mediating between efficient command recognition and privacy protection
2Adaptability or versatility
If user speech is transmitted to third-party services for processing, then the device can leverage external AI capabilities, but user control over sensitive information transmission is reduced
Solution Approach 1:
The system implements feedback by presenting identified SPI masking options to users and receiving their confirmation or modifications before transmission. This feedback loop ensures users maintain control over their sensitive information while still leveraging external AI capabilities for command processing
3Reliability
If speech recognition devices store user speech for quality assurance, then service quality can be monitored and improved, but stored speech becomes susceptible to eavesdropping and external leaks
Solution Approach 1:
The system extracts and removes sensitive personal information from user speech before transmission or storage. By taking out SPI through classification and masking, the system enables quality assurance storage of sanitized data while eliminating the security risks associated with storing raw speech containing sensitive information
4Object-affected harmful factors
If the system masks portions of user speech before transmission, then SPI protection is improved, but the quantity of information transmitted to services is reduced
Solution Approach 1:
The system applies local quality by selectively masking only the portions of speech that contain sensitive personal information while preserving the essential command information. This targeted approach protects SPI without unnecessarily reducing the functional information needed for command execution
Data Source
AI summary
A method, a computer program product, and a system for protecting SPI in spoken commands that include identifying sensitive personal information (SPI) included in unmasked portions of at least one classified command and following a user privacy instruction before transmitting the unmasked portions that include the identified SPI to a third-party service.