Target Speaker Identification in Multi-Speaker Voice Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voiceprint identification technologies struggle to accurately identify a target speaker in complex scenarios, such as multiple-speaker environments, and often require predefined keywords, limiting their effectiveness in securing electronic devices.
Innovation Solution
A neural-network-based method that includes a first network for extracting speaker features and determining the utterance scenario, followed by a second network for generating a final speaker feature, using a splicing layer, convolution layers, and attentive statistics pooling to enhance accuracy in both single- and multiple-speaker scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional voiceprint identification technology is used, then the system is simple to implement, but it cannot accurately identify target speakers in multiple-speaker environments
Solution Approach 1:
The voice signal is segmented into multiple speaker components using a speaker extraction network. The system divides the mixed voice signal into individual speaker embeddings, allowing accurate identification of target speakers even in multiple-speaker environments. This segmentation approach transforms the complex mixed signal into separable components that can be individually verified.
Solution Approach 2:
A speaker extraction network serves as an intermediary component between the input voice signal and the voiceprint verification system. This intermediary processes the mixed voice signal to extract and separate individual speaker features, enabling accurate target speaker identification without requiring the entire system to be fundamentally redesigned.
2Reliability
If voiceprint verification requires predefined keywords, then security can be enhanced, but the ease of operation decreases
Solution Approach 1:
The system performs automatic speaker extraction and verification without requiring users to provide predefined keywords or additional information. The speaker extraction network automatically identifies and separates target speaker voices from mixed signals, and the verification system automatically compares extracted features against registered templates, providing secure authentication through natural voice input alone.
3Adaptability or versatility
If speaker features are extracted from mixed voice signals, then the system can handle complex scenarios, but the measurement precision decreases
Solution Approach 1:
The system applies segmentation by separating mixed voice signals into individual speaker components through the speaker extraction network. This segmentation enables the system to handle multiple-speaker scenarios while maintaining extraction accuracy by processing each speaker's features separately rather than attempting to analyze the mixed signal as a whole.
Solution Approach 2:
The speaker extraction network performs preliminary processing of the mixed voice signal before verification occurs. By pre-separating and extracting individual speaker embeddings from the mixed signal, the system prepares clean, separated features for subsequent verification, ensuring high measurement precision even in complex multiple-speaker scenarios.
Data Source
AI summary
A processor-implemented method includes: extracting a target speaker voice feature based on an input voice of a target speaker; determining an utterance scenario of the input voice based on the target speaker voice feature; generating a final target speaker voice feature based on the determined utterance scenario; and determining whether the target speaker corresponds to a user based on the final target speaker voice feature and a final user voice feature, wherein the determined utterance scenario comprises either one of a single-speaker scenario and a multiple-speaker scenario.