双向交互机器人的语言自动生成仿真
|
摘要
针对当前有关语言文本检索方法存在的准确性差、查全率低问题,提出基于TF-IDF的双向交互机器人的语言自动生成方法。将语音特征曲线划分为小块,并在小块中调节语音特征曲线,将调节过的特征峰值进行归一化,实现特征曲线峰值配准与提取。将提取到的语音代入去噪环节,对语音中的噪声进行线性预测,并估计噪声分布,计算感知滤波器频域增益方程,将经傅里叶变换的语音噪声代入方程,获取去噪之后的纯净语音信号。基于语音预处理,利用TF-IDF相乘代表一个词或者短语权重,依据所有语句中出现的全部词构建整段话的矢量,将检索结果按照相似程度从大到小排序,将与检索词相似程度最高的当作机器人回复人类语音聊天的最佳选择,完成双向交互机器人的语言自动生成与回应。实验结果表明,所提方法检索准确率高,查全性能优于当前研究成果。上述方法运行精准度高,具有鲁棒性。
At present,the retrieval method about language text had low accuracy and low recall rate. Therefore,a method of language automatic generation for two-way interactive robot based on TF-IDF was proposed. The phonetic feature curve was divided into small blocks. In the small blocks,the phonetic feature curve was adjusted. The adjusted feature peak was normalized to achieve peak registration and extraction of feature curve. The extracted speech sound was substituted into the noise reduction and the noise in speech sound was predicted linearly. The noise distribution was estimated. The frequency domain gain equation of perceptual filter was calculated. In addition,the speech noise after Fourier transform was introduced into the equation to obtain the pure speech signal after noise reduction. Based on speech pretreatment,TF-IDF multiplication was used to represent the weight of word or phrase. According to all the words appearing in all the sentences,the vectors of whole paragraph were constructed. According to the similarity degree,the retrieval results were sorted from large to small. The retrieval result with the highest similarity degree to retrieval word was regarded as the best choice that the robot replied to human voice chat. Thus,automatic language generation and response of two-way interactive robot were completed. Simulation results show that the proposed method has higher retrieval accuracy and better recall performance. Meanwhile,this method has high running accuracy and robustness.
At present,the retrieval method about language text had low accuracy and low recall rate. Therefore,a method of language automatic generation for two-way interactive robot based on TF-IDF was proposed. The phonetic feature curve was divided into small blocks. In the small blocks,the phonetic feature curve was adjusted. The adjusted feature peak was normalized to achieve peak registration and extraction of feature curve. The extracted speech sound was substituted into the noise reduction and the noise in speech sound was predicted linearly. The noise distribution was estimated. The frequency domain gain equation of perceptual filter was calculated. In addition,the speech noise after Fourier transform was introduced into the equation to obtain the pure speech signal after noise reduction. Based on speech pretreatment,TF-IDF multiplication was used to represent the weight of word or phrase. According to all the words appearing in all the sentences,the vectors of whole paragraph were constructed. According to the similarity degree,the retrieval results were sorted from large to small. The retrieval result with the highest similarity degree to retrieval word was regarded as the best choice that the robot replied to human voice chat. Thus,automatic language generation and response of two-way interactive robot were completed. Simulation results show that the proposed method has higher retrieval accuracy and better recall performance. Meanwhile,this method has high running accuracy and robustness.
引文
[1] 颛悦,熊锦华,马宏远,等.一种支持混合语言的并行查询纠错方法[J].中文信息学报,2016,30(2):99-106.
[2] 司莉,张孝天.多语言知识组织系统的互操作项目调查及研究[J].情报科学,2016,34(9):3-6.
[3] 钟艳如,梁毅芳许本胜,等.基于网络本体语言的三维计算机辅助设计主模型相似性计算方法[J].计算机应用,2016,36(6):1599-1604.
[4] 佚名.多移动机器人系统协调控制算法设计[J].电子设计工程,2018,26(24):23-26.
[5] 鲍玉来,白淑霞,飞龙,等.汉蒙跨语言检索系统设计与实现[J].情报理论与实践,2017,40(4):128-132.
[6] 黄名选.基于加权关联模式挖掘的越英跨语言查询扩展[J].情报学报,2017,36(3):307-318.
[7] 韩朝,苗夺谦,任福继,等.基于粗糙集知识发现的开放领域中文问答检索[J].计算机研究与发展,2018,55(5):958-967.
[8] 袁庆升,张冬明,靳国庆,等.视频检索中图像信息量度量[J].通信学报,2016,37(2):80-87.
[9] 苏依拉,窦保媛,吉亚图.基于本体的蒙古语灾害信息检索模型[J].北京工业大学学报,2016,42(7):1017-1023.
[10] 刘媛.音乐库中检索特征音调歌曲智能识别仿真[J].计算机仿真,2017,34(8):356-359.
[11] 肖宝,李璞蒋运承.混合词汇特征和LDA的语义相关度计算方法[J].计算机工程与应用,2017,53(12):152-157.
[2] 司莉,张孝天.多语言知识组织系统的互操作项目调查及研究[J].情报科学,2016,34(9):3-6.
[3] 钟艳如,梁毅芳许本胜,等.基于网络本体语言的三维计算机辅助设计主模型相似性计算方法[J].计算机应用,2016,36(6):1599-1604.
[4] 佚名.多移动机器人系统协调控制算法设计[J].电子设计工程,2018,26(24):23-26.
[5] 鲍玉来,白淑霞,飞龙,等.汉蒙跨语言检索系统设计与实现[J].情报理论与实践,2017,40(4):128-132.
[6] 黄名选.基于加权关联模式挖掘的越英跨语言查询扩展[J].情报学报,2017,36(3):307-318.
[7] 韩朝,苗夺谦,任福继,等.基于粗糙集知识发现的开放领域中文问答检索[J].计算机研究与发展,2018,55(5):958-967.
[8] 袁庆升,张冬明,靳国庆,等.视频检索中图像信息量度量[J].通信学报,2016,37(2):80-87.
[9] 苏依拉,窦保媛,吉亚图.基于本体的蒙古语灾害信息检索模型[J].北京工业大学学报,2016,42(7):1017-1023.
[10] 刘媛.音乐库中检索特征音调歌曲智能识别仿真[J].计算机仿真,2017,34(8):356-359.
[11] 肖宝,李璞蒋运承.混合词汇特征和LDA的语义相关度计算方法[J].计算机工程与应用,2017,53(12):152-157.