基于自然语言处理的声纹优化识别算法研究

摘要

研究人的声纹的准确识别问题.声音特性没有一个统一的、不可变的标准,人的声音容易受到外界的干扰,声音的声纹美尔频率倒谱系数特征各项属性很容易发生变化.现有算法多半以声纹美尔频率倒谱系数为基础,识别效果容易受环境噪声、语音变异等因素的影响,造成声纹的干扰性变化,造成识别的精度不高.为此提出了一种基于语义特征和美尔频率倒谱系数特征相结合的声纹识别算法.利用MFCC准确提取语音中的频率特征,转化成自然语言环境下的语义特征,由于语义特征不受客观因素影响,减少了噪声信号对语音信号的影响,实现对语音特征的准确识别.实验表明,利用改进算法实现了差异化车辆图像的正确识别,提高了识别的准确度.%The problem of the accurate recognition of human voiceprint was researched in this paper.A voiceprint recognition algorithm was put forward based on semantic features and the United States Mel frequency cepstral coefficients features.Using the MFCC,the frequency characteristics of the speech were extracted accurately,then changed into the semantic features of the natural language environment.Because the semantic features are not affected by objective factors,so the influence of the noise signal on the speech signal was reduced,realizing the accurate recognition for speech characteristics.Experimental results show that the correct recognition of differentiated vehicle images was realized by using the improved algorithm and the accuracy of the recognition was also improved.