基于改进MFCC特征提取和DNN网络的机器人语音识别方法研究  

Research on Robot Speech Recognition Method Based on Improved MFCC Feature Extraction and DNN Network

作　　者：秦垲忻 王炜昕 王砚生 QIN Kaixin;WANG Weixin;WANG Yansheng(Kunming Putonghua Training and Testing Center,Kunming 650000,China;Kunming Network Security Emergency Command Center,Kunming 650500,China;School of Energy and Environment Science,Yunnan Normal University,Kunming 650500,China)

机构地区：[1]昆明市普通话培训测试中心,昆明650000 [2]昆明市网络安全应急指挥中心,昆明650500 [3]云南师范大学能源与环境科学学院,昆明650500

出　　处：《计算机测量与控制》2025年第2期246-253,共8页Computer Measurement &Control

基　　金：2022年云南省哲学社会科学规划项目(YB2022085);2024年全国教育规划青年课题(EHA210438)。

摘　　要：为了实现机器人语音控制,并避免环境噪音的干扰,研究提出了基于改进MFCC特征提取和深度神经网络的机器人语音控制指令识别方法;该方法利用线性判别分析、最大似然线性变换和说话人自适应变换对MFCC特征进行处理,获得了新的语音特征;同时通过深度玻尔兹曼机对声学模型进行了改进,并利用深度神经网络和谐波增强技术构建了语音增强方法;实验结果显示,研究提出的基于改进Mel频率倒谱系数特征能显著降低语音识别的字错误率,通过辅以改进深度神经网络-隐马尔科夫模型能进一步降低字错误率;在20 dB条件下,该特征和改进深度神经网络-隐马尔科夫模型的平均字错误率分别为24.9%和22.1%,均低于其他方法;上述结果表明,研究提出的语音识别方法能实现带噪声语音的准确识别,提高机器人的语音控制指令识别能力。To achieve robot voice control and avoid environmental noise interference,a robot voice control instruction recognition method based on improved Mel frequency cepstrum coefficient(MFCC)feature extraction and deep neural network(DNN)is proposed.This method utilizes linear discriminant analysis,maximum likelihood linear transformation,and speaker adaptive transformation to process MFCC features and obtain new speech features.Meanwhile,the acoustic model is improved using the deep Boltzmann machine,and deep neural networks and harmonic enhancement techniques are used to construct the speech enhancement method.Experimental results show that the proposed feature based on improved Mel frequency cepstral coefficients can significantly reduce the error rate in speech recognition,and further reduce the error rate by using an improved deep neural network hidden Markov model.Under the condition of 20 dB,the average error rates of this feature and improved deep neural network hidden Markov model are 24.9%and 22.1%,respectively,which are both lower than those of other methods.The above results indicate that the proposed speech recognition method can achieve accurate recognition of noisy speech and improve the ability of robot speech control command recognition.

关 键 词：语音识别 语音增强 声学模型 MFCC特征 DNN 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]