Gender-Specific Voiceprint Authentication Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current voiceprint recognition training processes use general models that result in low accuracy for distinguishing between genders.
Innovation Solution
A method and device utilizing a gender-mixed voiceprint baseline system based on Deep Neural Networks (DNN) to extract feature vectors, train gender classifiers, and develop separate DNN models for different genders, along with uniform background models and linear probability discriminant analysis models to enhance voiceprint authentication accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a general model is used for voiceprint training and recognition, then the training process is simple, but the accuracy of gender distinction is low
Solution Approach 1:
The patent divides the voiceprint recognition system into gender-specific segments by training separate Deep Neural Network models for male and female voices. The system first extracts features from speech segments, then uses a gender classifier to distinguish between male and female voices, and finally applies gender-specific DNN models for recognition. This segmentation approach directly addresses the low accuracy of general models in distinguishing genders while maintaining manageable complexity through modular model design.
2Reliability
If separate DNN models are trained for different genders, then the voiceprint authentication accuracy is improved, but the training time and computational resources increase
Solution Approach 1:
The patent implements preliminary gender classification before applying gender-specific DNN models. By first extracting features from speech segments and using a gender classifier to identify the speaker's gender, the system prepares the data in advance for the appropriate gender-specific model. This preliminary action optimizes the training and recognition process by ensuring that each speech segment is processed by the most suitable model, thereby improving authentication accuracy while managing training time efficiently through targeted model application.
Data Source
AI summary
The present disclosure provides a method and a device for processing voiceprint authentication. The method includes: extracting a first feature vector for each first speech segment of a training set by a gender-mixed voiceprint baseline system based on Deep Neural Network; training a gender classifier according to the first feature vector for each first speech segment and a pre-labeled first gender label of each first speech segment; training Deep Neural Network models for different genders respectively according to speech data of different genders of the training set; and training uniform background models, feature vector extracting models and linear probability discriminant analysis models for different genders respectively according to the Deep Neural Network models for different genders and the speech data of different genders of the training set. A voiceprint authentication processing model for gender distinguishing is built, thus improving the efficiency and accuracy of voiceprint authentication.