摘要
针对传统声纹识别方法在实际应用场景中跨设备情况下声纹识别性能较差的问题,提出了一种基于深度学习的跨设备声纹识别方法,采用了卷积循环网络的模型架构,在声纹注册阶段录制多段语音进行声纹特征的拟合建模,在识别阶段使用了切片降噪方式提取音频中的语音信息,在设备端使用了DSP芯片支持的双麦克采集现场声音。实验结果表明,在跨设备声纹识别条件下,本文提出的声纹识别方法识别准确率高于目前主流的方法,达到80%。
Aiming at the problem that the performance of voiceprint recognition is poor under the condition of cross-device voiceprint recognition in the actual application scenario of traditional voiceprint recognition method,a cross-device voiceprint recognition method based on deep learning is proposed.The model architecture of convolutional circular network is adopted,multi-segment speech is recorded in the voiceprint registration stage for fitting modeling of voiceprint features.The voiceprint recognition stage is used to extract audio speech information in the voiceprint recognition stage,and the dual microphone supported by the DSP chip is used to collect live sound on the device side.The test results show that the recognition accuracy of the voiceprint recognition method proposed in this paper is better than the current mainstream voiceprint recognition method under the condition of cross-device voiceprint recognition,which can achieve 80%voiceprint recognition accuracy.
作者
李伟
王鹏程
钟骁
喇二车龙补
Li Wei;Wang Pengcheng;Zhong Xiao;La Erchelongbu(Beijing Tongfanghuachuang Technology Co.,Ltd.,Beijing 100083,China;Tsinghua University(Department of Precision Instrument)and Tongfang Industrial Intelligent Microsystem Technology Joint Laboratory)
出处
《单片机与嵌入式系统应用》
2022年第12期16-19,共4页
Microcontrollers & Embedded Systems
关键词
卷积循环网络
深度学习
跨设备声纹识别
convolutional recurrent network
deep learning
cross-device voiceprint recognition
