Voiceprint Recognition Reducing Information Loss via Tail-Biting Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speaker recognition models face high information losses and system complexity due to the need for channel noise fitting processing and artificial feature engineering, which complicates data processing and loses information during feature extraction.
Innovation Solution
A method involving tail-biting convolution processing, discrete Fourier transform, and deep neural networks with a cosine similarity matrix loss function and minimum mean square error matrix loss function is used to segment and process original speech data, reducing noise and simplifying the system structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If channel noise fitting processing is performed on vector data to obtain fitted data, then speaker recognition accuracy is improved, but system complexity increases
Solution Approach 1:
The patent extracts and eliminates channel noise information from the speech data through the noise elimination module. By separating and removing the harmful channel noise component from the feature representation, the system achieves accurate speaker recognition without requiring complex channel noise fitting processing, thus resolving the contradiction between accuracy and complexity
Solution Approach 2:
The patent introduces an intermediate channel noise information representation that serves as a mediator between the raw speech data and the final speaker recognition decision. This intermediate representation allows the system to account for channel effects without requiring full noise fitting processing, simplifying the overall system while maintaining accuracy
2Loss of information
If artificial feature engineering is performed on original speech data, then feature information is extracted, but information loss occurs
Solution Approach 1:
The patent employs a universal deep neural network-based feature extraction framework that simultaneously performs multiple functions: automatic feature learning, channel noise estimation, and speaker feature extraction. This multi-functional approach eliminates the need for separate artificial feature engineering steps, reducing information loss while managing complexity
Solution Approach 2:
The patent replaces traditional mechanical artificial feature engineering processes with a data-driven deep learning system. The neural network automatically learns optimal feature representations from raw speech data, substituting manual feature design and eliminating information loss associated with manual feature extraction limitations
3Device complexity
If back-end reprocessing of calculation and analysis is performed on fitted data, then speaker recognition is achieved, but data processing complexity increases
Solution Approach 1:
The patent merges the feature extraction, noise elimination, and speaker recognition processes into an integrated deep neural network architecture. By combining these previously separate processing stages into a unified end-to-end system, the patent reduces data processing complexity while maintaining or improving processing efficiency through optimized computational flow
Data Source
AI summary
A method for voiceprint recognition of an original speech is used to reduce information losses and system complexity of a model for data recognition of a speaker's original speech. The method includes: obtaining original speech data, and segmenting the original speech data based on a preset time length to obtain segmented speech data; performing tail-biting convolution processing and discrete Fourier transform on the segmented speech data through a preset convolution filter bank to obtain voiceprint feature data; pooling the voiceprint feature data through a preset deep neural network to obtain a target voiceprint feature; performing embedded vector transformation on the target voiceprint feature to obtain corresponding voiceprint feature vectors; and performing calculation on the voiceprint feature vectors through a preset loss function to obtain target voiceprint data, where the loss function includes a cosine similarity matrix loss function and a minimum mean square error matrix loss function.