摘要
随着语音技术的发展,越来越多语音处理系统尝试应用于现实生活。然而在实际场景中,噪声干扰是一个影响语音识别等任务准确率的重要因素。为了克服噪声问题并提升性能,需设计语音分离或增强模块。文中通过结合波束成形与神经网络设计了在毫米半径麦克风阵列场景下的语音分离系统并在语音识别任务上进行了测试。实验显示文中设计的系统对语音识别准确率有一定帮助。该方法可以应用于设备空间受限的场景中以提高性能。
With the development of digital speech process technology,more and more audio systems have been applied in real life.However,in real scenes,noise is a fatal factor that will harm the performance of the system for tasks like Automatic Speech Recognition(ASR).In order to overcome noise affair and improve the system performance,a speech separation or enhancement module is reqired.In this paper,a speech separation system for a millimeter-level radius microphone array is designed by integrating beamforming methods and neural networks,and the system is tested on ASR tasks.The experiments show the proposed system helps the ASR system in recognition accuracy,and the system can be applied in a space-limited device scene.
作者
周祜旸
刘戈
方向忠
ZHOU Hu-yang;LIU Ge;FANG Xiang-zhong(School of Electronic Information and Electrical Engineering,Shanghai Jiaotong University,Shanghai 200240,China)
出处
《信息技术》
2023年第8期94-100,106,共8页
Information Technology
关键词
语音分离
语音增强
波束成形
差分麦克风阵列
语音识别
speech separation
speech enhancement
beamforming
difference microphone array
Auto-matic Speech Recognition
