Voice Authentication Vector Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for providing user-specific services or access, such as voice-based authentication, face challenges in balancing speed, ease of use, security, and accuracy, particularly in scenarios like voice-first devices and shared accounts, where traditional username-password combinations are not ideal.

Innovation Solution

A system that uses a vector space model to analyze utterances, segmenting them into activation triggers, commands, and parameters, and represents these in vector spaces for comparison against registered user models to determine user identity and provide access to associated accounts, enhancing security and convenience.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If voice-based authentication is used to improve speed and ease of use, then authentication time is reduced and user convenience is improved, but security and accuracy deteriorate due to inability to distinguish between different users with similar voices

Engineering Contradiction:
Improveease of useVSAvoidsecurity
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The patent segments the utterance into multiple portions (activation trigger portion, command portion, parameter portion) and creates separate vector space models for each segment. This allows the system to authenticate users based on multiple independent segments, making it more difficult for impostors to succeed even if they can replicate one segment of speech.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms voice authentication from a single-dimensional comparison to a multi-dimensional vector space comparison. By representing utterance portions as vectors in high-dimensional space and comparing them against stored user profiles, the system achieves more accurate and secure authentication while maintaining ease of use.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Reliability

If traditional username-password authentication is used to improve security, then authentication accuracy is maintained, but ease of use and speed deteriorate especially on devices without keyboards

Engineering Contradiction:
ImprovesecurityVSAvoidease of use
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The patent replaces the mechanical keyboard input system with an acoustic field-based voice recognition system. Users interact with the device through speech commands, eliminating the need for physical keyboard input while maintaining secure authentication through voiceprint analysis and vector space comparison.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces an intermediary vector space model layer between the voice input and user authentication. This intermediary representation allows the system to securely compare voice characteristics without requiring direct username-password input, enabling both security and ease of use.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If speaker verification is performed on entire utterances to improve accuracy, then authentication reliability is improved, but processing time increases and speed deteriorates

Engineering Contradiction:
ImproveaccuracyVSAvoidspeed
Core Design Contradiction:
ReliabilityVSSpeed

Solution Approach 1:

The patent divides the utterance into smaller portions (activation trigger, command, parameter) and creates corresponding vector space models for each. This segmentation allows the system to perform faster, more targeted comparisons on smaller data chunks rather than processing entire utterances at once, improving both speed and accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary action by pre-computing and storing vector space models for common utterance portions during system setup. When authentication is needed, the system can quickly compare new utterance portions against these pre-prepared models, significantly reducing processing time while maintaining high accuracy.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3786945B1Voice-based authentication
Publication Date: 2023.07.19 SPOTIFY
  • EP3786945B1 patent drawingFigure 1
  • EP3786945B1 patent drawingFigure 2
  • EP3786945B1 patent drawingFigure 3

AI summary

Voice-based authentication can include methods, systems, devices, and computer program products for providing user-specific services or access based at least in part on an utterance. In one method, an audio clip having an utterance is obtained. The utterance has an activation trigger portion and a command portion. A first distance between a vector representation of the activation trigger portion and a registered activation trigger vector is determined; and a second distance between a vector representation of the command portion and a registered command vector is determined. Responsive to the first distance satisfying a first distance threshold, and the second distance satisfying a second distance threshold, access is provided to a service associated with a registered user.