Voice Authentication Using Random Phrases and Utterance Cues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing identity authentication methods are vulnerable to deception using recorded speech data and synthesized speech generated by advanced speech synthesis technology.
Innovation Solution
An identity authentication device generates a unique authentication sentence with specified utterance methods, including speed, inflection, and emotional expression, and requires input within a deadline, using machine learning to verify voiceprint, wording, and utterance method matches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If speech authentication is performed using voice recognition, then user authentication can be conducted, but the system becomes vulnerable to deception by recorded speech data and synthesized audio
Solution Approach 1:
The patent applies dynamics by making the authentication phrase random and changing it for each authentication attempt. The system dynamically adjusts the phrase based on timing information and presents it only for a limited duration, preventing the use of pre-recorded or synthesized responses. This dynamic approach transforms the static authentication process into an adaptive one that resists replay attacks.
Solution Approach 2:
The patent changes multiple parameters of the authentication process: the authentication phrase is randomly generated, the timing window is precisely controlled, and the phrase is presented only for a limited time. These parameter changes make it impossible for attackers to use recorded or synthesized speech, as the system continuously varies the authentication parameters.
2Reliability
If a random authentication phrase is presented to eliminate recorded speech deception, then security against recorded data is improved, but the system cannot handle synthesized speech with read authentication phrases
Solution Approach 1:
The system incorporates feedback mechanisms by detecting the actual utterance timing and comparing it against the expected timing window. The system provides feedback through the timing validation process, accepting only responses within the precise time window and rejecting others, thereby countering synthesized speech that cannot replicate the exact timing characteristics.
Solution Approach 2:
The patent replaces purely acoustic-mechanical speech recognition with a system that incorporates temporal and contextual analysis. By adding timing dimensions and random phrase generation, the system moves beyond simple voice pattern matching to a multi-parameter verification process that distinguishes human speech from synthesized or recorded versions.
3Reliability
If an authentication deadline is set to prevent synthesized speech, then security against synthesis is improved, but the user experience may be negatively affected
Solution Approach 1:
The system performs preliminary action by pre-calculating the authentication deadline based on the expected speaking time. The deadline is set to approximately twice the time required to finish reading the authentication sentence at the presented utterance speed. This preliminary timing setup ensures that legitimate users have sufficient time to respond while synthesized speech, which is faster, cannot complete the authentication within the window.
Solution Approach 2:
The patent applies partial action by setting the authentication deadline to be approximately twice the reading time, which provides a buffer that accommodates normal variations in user speaking speed. This excessive time allocation ensures that legitimate users are not unfairly rejected while still preventing synthesized speech that operates at much faster speeds.
Data Source
AI summary
An identity authentication device presents an authentication sentence generated at a time of an authentication by a sentence generation unit to a user using an authentication speech presentation unit, together with an utterance method set by an utterance method designation unit, and performs authentication based on a voiceprint, wording, and an utterance method detected from speech data input to a speech input unit using a voiceprint authentication unit, a speech authentication unit, an utterance method determination unit, and an authentication result determination unit.