·人工智能及识别技术·1000ATP18用于人形机器人的声音降噪方法.pdf

第 35 卷 第 5 期 计 算 机 工 程 2009 年 3 月 Vol.35 No.5 Computer Engineering March 2009 ·人工智能及识别技术· 文章编号：1000—3428(2009)05—0191—03 文献标识码：A 中图分类号：TP18 用于人形机器人的声音降噪方法 林剑峰，冯 瑞 (复旦大学计算机科学与工程系肢体化智能实验室，上海 200433) 摘 要：用于人形机器人的实时多通道降噪方法使机器人能够在实际应用中精确地进行后续的音频分类或语音识别。引入人形机器人具有 类似人体的头相关传输函数的性质，避免一般方法中在估计声传播信道模型时只考虑声达时延，而忽略声散射和共振影响的局限性。该方 法通过利用麦克风阵列阵元之间的声音传输信道的差异性，削弱参考噪声和目标信号之间的相关性，从而提升对消噪声后输出信号的信噪 比。仿真表明，对目标信号的信噪比提升达到15 dB。 关键词：人形机器人；声音降噪；麦克风阵列；头相关传输函数 Noise Reduction Method for Humanoid Robot LIN Jian-feng, FENG Rui (Embodied Intelligence Lab, Department of Computer Science and Engineering, Fudan University, Shanghai 200433) 【Abstract 】A real time, multichannel, humanoid robot ready noise reduction method makes practical precise audio classification and speech recognition possible. Alhough scattering of sound plays an important role on humanoid robot, most conventional methods consider time delay as the only main factor for transferring model of sound propagation channel between microphones. By using characteristics of human like Head Related Transfer Function(HRTF) system, this method avoids the limitation of lackness of scattering effect. Simulations show that this method is able to improve desired sound’s SNR by 15 dB. 【Key words 】humanoid robot; noise reduction; microphone array; Head Related Transfer Function(HRTF) ˆ 1 概述 获得最优的噪声估计b(n) ，从而进行噪声消减。 + +