Real-Time Media Anonymization for Bias-Free Video Interactions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing remote conferencing tools exacerbate disparities and cognitive biases in media-based interactions due to user physical and audio characteristics, necessitating a solution to anonymize users and reduce bias.
Innovation Solution
A system and method that alters video and audio data in real-time to anonymize users by replacing identifying features with neutral or anonymizing synonyms, using an augmented reality overlay module to generate a virtual representation of the user during interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If remote conferencing tools are used to enable virtual interactions, then communication accessibility is improved, but cognitive biases and disparities are amplified
Solution Approach 1:
The patent introduces an automated assistant as an intermediary between users and the conferencing system. This assistant processes audio and video data, applies anonymization techniques, and filters out identifying characteristics that trigger cognitive biases, while still enabling effective communication
Solution Approach 2:
The system creates anonymized copies of user representations (audio and video) that preserve communication functionality while removing identifying features. These synthetic representations allow interactions to proceed without exposing users to bias based on their actual characteristics
2Reliability
If user characteristics are visible and audible in virtual interactions, then personal connection is enhanced, but bias and discrimination increase
Solution Approach 1:
The system extracts and removes identifying characteristics from audio and video streams while preserving the essential communication content. Natural language processing identifies and filters words related to protected characteristics, and audio processing removes vocal tone indicators that reveal user identity
Solution Approach 2:
The anonymization process selectively modifies specific aspects of the communication stream (certain words, vocal tones, visual identifiers) while leaving other aspects intact, applying different levels of processing to different parts of the data based on their potential to trigger bias
3Object-affected harmful factors
If natural language processing is applied to filter audio data, then speech anonymization is achieved, but processing complexity increases
Solution Approach 1:
The system performs preliminary processing of audio data by converting speech to text before applying natural language filtering. This preliminary transcription step enables more efficient text-based analysis and filtering of identifying words and phrases
Solution Approach 2:
The natural language processing system continuously monitors the communication stream, identifies words related to protected characteristics, and dynamically adjusts filtering in real-time based on the context and content of the conversation
4Object-affected harmful factors
If video data is altered to anonymize facial information, then visual bias is reduced, but information loss occurs
Solution Approach 1:
The system creates anonymized video copies where facial features are obscured or replaced with neutral representations. These synthetic video streams preserve the visual presence and basic expressions needed for communication while eliminating identifying facial characteristics
Data Source
AI summary
Disclosed herein are a computing device, method, and computer-readable medium embodiments for altering video and/or audio data during a media-based interaction to protect anonymity of one or more users in the media-based interaction for the purpose of removing bias from the media-based interaction. In some aspects, video and audio data including facial information and speech data of a user may be streamed during the media-based interaction between client devices. The video and audio data may be altered in real-time during the media-based interaction to change the visual and/or audio aspects of one or more users during the media-based interaction. This alterations to the video and audio data may be based on identifying visual and audio features from a list of one or more visual and audio identifiers associated with age, sex, and/or gender. The alteration results in a new virtual representation presented during the interaction that anonymizes the user.