Shared-Microphone Voice Attribution With Candidate Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing communication technologies lack the ability to intelligently determine the identity of users sharing a single audio source, particularly in teleconferencing and meeting scenarios, and fail to provide visual indicators for hearing-impaired users, leading to inefficiencies in computing resource consumption and user interface deficiencies.
Innovation Solution
Implementing a system that utilizes Gaussian Mixture Models and Hidden Markov Models to attribute voice utterances to specific users, combined with user relationship objects to generate participant candidates, reducing unnecessary comparisons and improving computing efficiency and accuracy by pruning potential users based on network proximity, and providing visual and audio indicators of the speaking user.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voice utterances are compared against all users in an organization to determine identity, then user identification accuracy may improve, but computing resource consumption increases significantly
Solution Approach 1:
The patent segments the organization's user base into participant candidates based on event context and user relationships. Instead of comparing against all users, the system divides the comparison set into relevant subsets (participants invited to the event, users with recent activity, etc.), thereby reducing computational load while maintaining identification accuracy for the actual speaker.
Solution Approach 2:
The system performs preliminary actions by pre-identifying participant candidates before voice comparison begins. User relationship objects are pre-computed and stored, allowing the system to quickly filter potential speakers based on event invitations and relationship data, eliminating the need for exhaustive comparisons against all organizational users.
2Reliability
If comprehensive user comparison is performed for voice attribution, then identification reliability improves, but processing time increases
Solution Approach 1:
The patent segments the user comparison process into multiple stages: first filtering by event participation status, then by relationship proximity, and finally by voice characteristics. This segmented approach maintains reliability by systematically evaluating relevant criteria while reducing overall processing time through early elimination of non-candidates.
Solution Approach 2:
The system performs partial comparison by focusing only on the most relevant participant candidates rather than all users. By applying the 80/20 rule, the system achieves sufficient identification reliability by comparing against the top N candidates with strongest contextual evidence, avoiding unnecessary comparisons that would increase processing time without significantly improving accuracy.
3Ease of operation
If visual indicators are added to identify speaking users, then user interface usability improves, but device complexity increases
Solution Approach 1:
The patent applies local quality by providing visual indicators specifically for the speaking user's identity information rather than redesigning the entire interface. The system locally enhances the user experience by displaying relevant identification data (name, avatar, location) in the context where the user needs it most - during audio communication - without adding complex functionality elsewhere in the system.
Solution Approach 2:
The system introduces visual indicators as an intermediary layer between the voice identification system and the user. Rather than directly displaying complex identification data structures, the system mediates this information through user-friendly visual representations (avatars, name tags, location indicators) that simplify interaction while maintaining access to comprehensive identification capabilities.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Computerized systems are provided for determining an identity of one or more users that use a same audio source, such as a microphone. The identity of one or more users that use a same audio source can based on generating a list of participant candidates who are likely to participate in an associated event, such as a meeting. For instance, embodiments can generate one or more network graphs of a meeting invitee any only voice input samples of the meeting invitee's N closest connections are compared to an utterance to determine the identity of the user associated with the utterance. One or more indicators that identify the users who are using the same audio source, as well as additional information or metadata associated with the identified user can be caused to be presented.