Biometric Speaker Identification via Clustering and Nearness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated speaker identification based on voice information has had limited success, necessitating an improved method to accurately identify individuals using biometric data.
Innovation Solution
A biometric speaker-identification apparatus that processes voice match scores through k-means clustering and nearness measurements in a P-dimensional hyperspace to select and order speaker-identity candidates, utilizing Euclidean distance and dot product similarity for effective matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional voice matching is used to identify speakers, then the process is simple, but the identification accuracy is limited
Solution Approach 1:
The patent segments the speaker identification process into multiple stages: initial voice matching to generate candidate templates, clustering to group similar candidates, and nearness measurement to identify the best match. This segmentation allows complex processing to be applied selectively to reduce accuracy limitations without requiring complete reprocessing of all templates.
Solution Approach 2:
The patent performs preliminary clustering and nearness measurements on candidate templates before final speaker identification. By pre-processing the candidate set through k-means clustering and eliminating unlikely candidates via nearness thresholds, the system prepares optimized candidate lists in advance, improving identification accuracy while managing computational complexity.
2Reliability
If all templates in the biometric corpus are evaluated, then complete coverage is achieved, but processing time increases
Solution Approach 1:
The patent extracts and evaluates only the most relevant candidate templates from the biometric corpus by using clustering to identify promising groups and applying nearness measurements to filter out unlikely candidates. This extraction approach maintains identification reliability by focusing on high-probability matches while significantly reducing processing time compared to evaluating all templates.
Solution Approach 2:
The patent applies partial action by evaluating a subset of templates rather than the complete corpus. Through clustering and nearness-based filtering, the system processes only the necessary portion of templates required to achieve reliable identification, eliminating redundant computations on clearly unrelated candidates.
3Measurement precision
If clustering and nearness measurements are applied, then candidate selection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the computational process into distinct phases: k-means clustering to group templates, nearness measurement to evaluate proximity to probe, and final candidate selection. This segmentation allows each computational task to be optimized independently and applied only where needed, improving candidate selection accuracy while managing overall computational complexity.
Solution Approach 2:
The patent performs clustering and nearness measurements as preliminary actions before final speaker identification. By pre-computing cluster assignments and nearness scores for candidate templates, the system prepares refined candidate lists that improve selection accuracy while avoiding the need for complex real-time computations during the actual identification process.
Data Source
AI summary
A biometric speaker-identification apparatus is disclosed that generates ordered speaker-identity candidates for a probe based on prototypes. Probe match scores are clustered, and templates that correspond to clusters having top M probe match scores are compared with the prototypes to obtain template-prototype match scores. The probe is also compared with the prototypes, and those templates corresponding to template-prototype match scores that are nearest to probe-prototype match scores are selected as speaker-identity candidates. The speaker-identity candidates are ordered based on their similarity to the probe.