Voice Authentication Model for User-Chosen Passphrase Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing authentication systems require multiple models for different passphrases and languages, increasing costs and training time, and force users to use the same passphrase, compromising security and convenience.
Innovation Solution
A machine learning model is trained using a dataset of audio signals containing voice biometric data and passphrase information, allowing users to choose their own passphrases, and enabling simultaneous verification of identity and passphrase matching through embeddings and triplet loss calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple TD-ASV models are trained for different passphrases, then authentication accuracy is improved, but training time and costs increase
Solution Approach 1:
The patent combines multiple passphrase-specific TD-ASV models into a single universal model that can handle multiple passphrases. The model uses a shared acoustic model and embedding layer that processes different passphrases through the same neural network architecture, eliminating the need to train separate models for each passphrase while maintaining authentication accuracy.
Solution Approach 2:
The patent creates a universal ASV model that functions across multiple passphrases and languages. The model is designed with language-agnostic features and can be trained on multilingual data, allowing a single model to serve multiple authentication scenarios that previously required separate specialized models.
2Reliability
If multiple ASR models are trained for different languages, then language-specific recognition accuracy is improved, but device complexity and costs increase
Solution Approach 1:
The patent develops a universal ASR model that can process multiple languages using a single model instance. The model incorporates multilingual training data and uses language-agnostic acoustic features, allowing it to recognize speech in different languages without requiring separate language-specific models, thus reducing device complexity while maintaining recognition accuracy.
3Ease of operation
If all users speak the same passphrase, then system simplicity is improved, but security is worsened
Solution Approach 1:
The patent implements passphrase-specific embedding layers within the universal model that allow each user's passphrase to be processed with specialized parameters. While the overall model structure remains simple and unified, local customization is achieved through user-specific embedding matrices that capture individual passphrase characteristics, providing both system simplicity and enhanced security.
4Reliability
If TI-ASV and ASR models are trained separately, then model specialization is improved, but training calculations and costs increase
Solution Approach 1:
The patent merges the TI-ASV and ASR training processes into a unified joint training framework. The model simultaneously optimizes both speaker verification and speech recognition objectives using a combined loss function that incorporates elements from both tasks, allowing the system to learn both capabilities in a single training process rather than requiring separate training runs.
Data Source
AI summary
A method of training a machine learning model to conduct authentication transactions is provided that includes the steps of obtaining, by an electronic device, a training dataset of audio signals. Each audio signal includes voice biometric data of a user and information for a passphrase spoken by the respective user and belongs to a same or different data class. Each data class includes a user identity and a passphrase identifier. Moreover, the method includes the steps of creating, using a machine learning model being trained, at least one embedding for each audio signal. The machine learning model includes parameters. Furthermore, the method includes calculating, by a machine learning algorithm using the embeddings, a loss, and updating parameters of the machine learning model based on the calculated loss. In response to determining criteria defining an end of training have been satisfied, deeming the machine learning model to be operable for use in simultaneously successfully verifying the identity of a user based on voice biometric data and verifying a passphrase spoken by the user matches a secret passphrase during authentication transactions.