深度卷积神经网络在多语种说话人识别中的应用OA
Application of Deep Convolutional Neural Network in Multilingual Speaker Recognition
针对复杂噪声环境下多语种说话人识别鲁棒性不足的问题,提出一种基于多分辨率卷积神经网络(MRCNN)的识别方法.通过设计不同卷积核尺寸的多层CNN架构,捕捉语音信号中不同频率范围的特征,解决复杂噪声环境下的鲁棒性问题.在Aurora2多条件噪声集上的实验结果表明,所提出的方法的平均识别准确率达到了61.2372％,较深度置信网络(DBN)提升了3.4500％,较传统CNN提升了0.5789％.所提出的方法为多语种说话人识别提供了一种高效且鲁棒的解决方案,在复杂语音环境下具有广阔应用前景.
Aiming at the issue of insufficient robustness of multilingual speaker recognition in complex noisy environments,this paper proposes a recognition method based on multi-resolution convolutional neural network(MRCNN).By designing the multi-layer CNN architecture with different convolution kernel sizes,the features of different frequency ranges in speech signals are captured and the robustness problem in complex noise environments is solved.The experimental results on the Aurora2 multi-condition noise set show that the average recognition accuracy of the proposed method reaches 61.2372％,which is 3.4500％higher than that of the deep belief network(DBN)and 0.5789％higher than that of the traditional CNN.The pro-posed method provides an efficient and robust solution for multilingual speaker recognition,which has broad application pros-pects in complex speech environments.
孙杰;杨治学
昌吉学院,信息工程学院,新疆,昌吉 831100昌吉学院,信息工程学院,新疆,昌吉 831100
信息技术与安全科学
深度卷积神经网络多分辨率卷积神经网络多语种说话人识别噪声鲁棒性
deep convolutional neural networkmulti-resolution convolutional neural networkmultilingual speaker recogni-tionnoise robustness
《微型电脑应用》 2026 (4)
6-8,3
新疆维吾尔自治区科技厅自然科学面上项目(2022D01C03)
评论