基于上下文敏感区块的模糊语音准确识别方法  

Accurate recognition of fuzzy speech based on context sensitive block

作　　者：全龙翔 阿不力克木·吾甫尔 马超 武江波 QUAN Long⁃xiang;Abulikemu·Wupuer;MA Chao;WU Jiang⁃bo(State Grid Xinjiang Electeic Power Research Institute CO.,LTD,Urumqi 830000,China)

机构地区：[1]国网新疆电力有限公司电力科学研究院

出　　处：《电子设计工程》2020年第1期32-35,44,共5页Electronic Design Engineering

基　　金：江苏省科技厅项目(CGYKJQQ00000019)

摘　　要：为对语音信号进行良性切分,实现有目的性的声源重组,提出一种基于上下文敏感区块的模糊语音准确识别方法。在区块组织的频谱特征中,确定模糊语音的Gabor滤波传输条件,并对Delta描述算子进行定向计算,完成上下文敏感区块模糊语音的特征参数分析。在此基础上,利用深度识别神经网络,对模糊语音的特征线索进行有效分离,并对其识别端点进行逐一排查,完成新型语音准确识别方法的构建。对比实验数据显示,与基础语音识别方法相比,基于上下文敏感区块的模糊语音准确识别方法既可将最大信号切分率提升至95%左右,也能保持声源信号的最大深度不超过4.50×10^-7μm,达到重组声源的目的。In order to segment speech signal benignly and achieve purposeful source reorganization,an accurate recognition method based on context-sensitive blocks for fuzzy speech is proposed. In the spectrum characteristics of the block organization,the Gabor filter transmission condition of the fuzzy speech is determined,and the Delta descriptor is calculated in orientation to complete the analysis of the characteristic parameters of the context-sensitive block fuzzy speech. On this basis,the deep recognition neural network is used to effectively separate the feature clues of the fuzzy speech,and the recognition endpoints are checked one by one to complete the construction of a new accurate speech recognition method. The experimental results show that compared with the basic speech recognition method,the context-sensitive block-based fuzzy speech recognition method can not only increase the maximum signal segmentation rate to about 95%,but also maintain the maximum depth of the source signal not more than 4.50*10^-7μm,so as to achieve the purpose of recombining the source.

关 键 词：敏感区块 模糊语音 频谱特征 GABOR滤波 Delta描述子 

分 类 号：TN912[电子电信—通信与信息系统]