基于一维卷积网络和自注意力机制的语种识别算法OA

Language recognition algorithms based on embedded vector representation can achieve excellent results in ideal environment.However,in real-world scenarios,such as dialect recognition in case analysis,performance is often compromised by noise.Current algorithms are not significantly robust for language recognition modeling in environment with low signal-to-noise ratio.In this paper,a language embedded vector representation algorithm based on self-attention mechanism and one-dimensional convo-lutional neural network is proposed.The algorithm uses one-dimensional convolutional neural network in-stead of the traditional time-delay neural network to fuse the time sequence information more efficiently at the audio frame level.Meanwhile,self-attention mechanism is introduced to enhance the model's ability to focus on specific language features,even in noisy environment,thereby enhancing the robustness of the model for language recognition in noisy environment.By testing the language audio data in a simulated noisy environment,the experimental results show that compared with the mainstream algorithms,the pro-posed algorithm achieves a 2.21%improvement in recognition accuracy for 10-second audio in noisy en-vironments.It also reduces the equal error rate by 0.24%,and the average detection cost by 0.20%.