Siamese Network Speaker Recognition With Attention Mechanism
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speaker recognition methods have low accuracy due to weak representation ability of features extracted, leading to ineffective identification of speakers in applications like classroom quality evaluation, where the duration and frequency of teacher and student interactions are critical.
Innovation Solution
A speaker recognition method utilizing a Siamese neural network for feature learning on audio feature sequences, combined with an attention mechanism-based machine learning model to enhance the accuracy of speaker identification by leveraging continuous information in audio files.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If common deep neural network models (CNN, RNN) are used for speaker recognition, then the system is simple to implement, but the accuracy of speaker recognition results is low due to weak representation ability of extracted features
Solution Approach 1:
The patent combines multiple neural network components (Siamese network for feature extraction, attention mechanism for weighting, and pooling layers for aggregation) to create a composite model architecture. This composite approach integrates the strengths of different network types to achieve superior speaker recognition accuracy while managing complexity through modular design
Solution Approach 2:
The patent introduces an attention mechanism as an intermediary component between feature extraction and final recognition. This attention mechanism weights and prioritizes different audio features, acting as a mediator that enhances the representation ability of extracted features without requiring complete redesign of the base neural network architecture
2Reliability
If audio files are segmented into multiple audio units for processing, then continuous voice signal information is preserved, but the processing complexity increases
Solution Approach 1:
The patent divides audio files into multiple audio units or segments, allowing the system to process and analyze continuous voice signals in manageable portions. This segmentation enables the extraction of temporal patterns and continuous information while maintaining computational feasibility through structured processing of divided segments
Solution Approach 2:
The patent combines multiple segmented audio units back together through pooling operations and attention mechanisms, merging the processed information from individual segments to form a comprehensive speaker representation. This merging process preserves continuous voice signal characteristics while leveraging the benefits of segmented processing
Data Source
AI summary
The present disclosure provides a speaker recognition method, an electronic device, and a storage medium. An implementation includes: segmenting the target audio file and the to-be-recognized audio file into a plurality of audio units respectively; extracting an audio feature from each of the audio units to obtain an audio feature sequence of the target audio file and an audio feature sequence of the to-be-recognized audio file; performing feature learning on the audio feature sequence of the target audio file and the audio feature sequence of the to-be-recognized audio file by using Siamese neural network, to obtain a feature vector corresponding to the target audio file and feature vectors respectively corresponding to the plurality of audio units in the to-be-recognized audio file; and recognizing, by using an attention mechanism-based machine learning model, the audio units belonging to the target speaker in the to-be-recognized audio file based on the feature vectors.