Role Separation Using Sound Source Position and Voiceprint Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice separation technologies face significant errors when distinguishing between speakers with similar voiceprint features, leading to incorrect role identification in applications like conference scenes.
Innovation Solution
A role separation method that utilizes sound source information to filter candidate positions, followed by calculating the similarity between voiceprint features to accurately determine the target role, incorporating both sound source position and voiceprint characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If voiceprint features alone are used for role separation, then the method is simple, but accuracy deteriorates when speakers have similar voiceprint features
Solution Approach 1:
The patent combines sound source position information and voiceprint features into a fused feature vector for role separation. The sound source position is encoded as a one-hot vector and concatenated with the voiceprint feature vector, creating a composite feature that leverages both spatial and acoustic characteristics to improve identification accuracy when speakers have similar voiceprint features.
Solution Approach 2:
The patent introduces a spatial dimension (sound source position) to the traditional voiceprint-based role separation. By adding positional information as an additional dimension to the feature space, the system can distinguish between speakers with similar voiceprint features based on their different spatial locations, effectively transforming a 1D voiceprint comparison into a 2D problem incorporating both voice and position.
2Reliability
If all positions are considered for voiceprint comparison, then comprehensive coverage is achieved, but computational load increases
Solution Approach 1:
The patent segments the comparison process into two stages: first filtering candidate positions based on sound source position information, then performing voiceprint comparison only on these filtered candidates. This segmentation reduces the number of comparisons needed while maintaining reliability, as the position-based filtering eliminates obviously incorrect candidates before the computationally intensive voiceprint matching.
Solution Approach 2:
The patent performs preliminary filtering of candidate positions using sound source position information before conducting the actual voiceprint comparison. This preliminary action narrows down the search space to only those positions that are acoustically plausible, thereby reducing the computational load of the subsequent voiceprint matching while ensuring that the correct position is not excluded.
Data Source
AI summary
Embodiments of the present application provide a role separation method, an electronic device, and a computer storage medium. The role separation method includes: acquiring sound source information of target voice data and a voiceprint feature of the target voice data; determining, according to the sound source information, at least one candidate position corresponding to a sound source position; calculating a similarity between a voiceprint feature of a role corresponding to the at least one candidate position and the voiceprint feature of the target voice data; and determining a target role corresponding to the target voice data according to the similarity. By means of the embodiments of the present application, the accuracy of the role separation is improved.