摘要
当机器人移动身体任何部位时,都会不可避免地产生自身噪声。这些自身噪声由身体关节或其他硬件设备如风扇等引起。由于自身噪声距离机器人麦克风较近,较目标声源更容易被获取。该文根据机器人自身噪声种类,提出了一种将谱减法、关节噪声模板减法、基于标注区域的倒谱均值减法以及多条件训练相结合的方法,从而估计和抑制自身噪声。一系列实验证明了所提出的方法可以有效地减少自身噪声影响,提高语音识别的鲁棒性。
Robots inevitably produce noise when they are moving any part of their body. Such noise is caused by the various body joint motors as well as the CPU cooling fans. Moreover, these noises are easily captured by the robots＇ microphones because they are closer to the microphones than the target speech source. This paper presents a de-noising method using the spectral subtraction, joint noise template substraction, labeled area cepstral mean substraction and multi-condition training to estimate and suppress robot noise. Tests show that this method significantly reduces the effect of robot noise which enhances the automatic speech recognition.
作者
王建荣
张句
路文焕
魏建国
党建武
WANG Jianrong Z.HANG Ju LU Wenhuan WEI Jianguo DANG Jianwu(School of Computer Science and Technology, Tianjin University, Tianjin 300072, China School of Computer Software, Tianjin University Tianjin 300072, China)
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2017年第2期153-157,共5页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(61471259
61304250
61573254)
关键词
机器人
语音识别
语音增强
robot
speech recognition
speech enhancement
