融合递归求逆滤波的机器人混合语音识别方法  

Simulation of Robot Hybrid Speech Recognition Based on Recursive Inverse Filtering

作　　者：李丽亚[1] 闫宏印[2] LI Li-ya;YAN Hong-yin(Taiyuan Institute of Technology,Taiyuan Shanxi 030008,China;Taiyuan University of Technology,Taiyuan Shanxi 030024,China)

机构地区：[1]太原工业学院,山西太原030008 [2]太原理工大学,山西太原030024

出　　处：《计算机仿真》2020年第8期277-280,290,共5页Computer Simulation

摘　　要：机器人语音识别需要满足混合噪声与非特定人等复杂场景的使用需求,现有方法难以达到全场景的准确识别。为此提出了融合递归求逆滤波的混合语音识别方法。方法根据机器人语音识别模型,分别对语音信号的预处理,特征提取,以及匹配识别三个阶段做了相应优化。在预处理阶段,设计了递归求逆算法,用于补偿原始语音中的小特征信号,同时采用加权向量,在递推过程中增强原始语音信号。在特征提取阶段,针对卷积噪声,边缘效应,以及基音干扰,设计了多次滤波器,用以避免语音特征提取时的各种干扰。最后的匹配识别阶段,利用提取的特征训练得到模板库,通过欧氏距离递推得到相似度,并根据阈值约束完成匹配识别。仿真结果表明,融合递归求逆滤波方法显著提高了混合语音信号的识别率与抗噪性能,能够更好的满足机器人复杂场景下的应用需求。Robot speech recognition needs to meet the needs of complex scenes such as mixed noise and non-specific person.It is difficult for existing methods to achieve accurate recognition of the whole scene.Therefore,the paper proposed a hybrid speech recognition method based on recursive inverse filtering.According to the speech recognition model of robot,this method optimizes the three stages of speech signal preprocessing,feature extraction and matching recognition.In the pre-processing stage,a recursive inverse algorithm was designed to compensate the small feature signals in the original speech.At the same time,the weighted vector was used to enhance the original speech signal in the recursive process.In the feature extraction stage,aiming at convolution noise,edge effect and pitch interference,a multiple filter was designed to avoid all kinds of interference in speech feature extraction.At the last stage of matching recognition,the template library was obtained by training the extracted features.Similarity was obtained by Euclidean distance recursion,and matching recognition was completed according to threshold constraints.The simulation results show that the fusion recursive inverse filtering method can significantly improve the recognition rate and anti-noise performance of mixed speech signals.It can better meet the application requirements of robots in complex scenarios.

关 键 词：机器人场景 递归求逆 多次滤波 欧氏距离 混合语音识别 

分 类 号：U491[交通运输工程—交通运输规划与管理]