摘要
说话人识别是信息技术和生物学的新一代身份验证方式,在说话人识别的研究中,特征参数的提取直接影响到识别系统最终的识别效率.通过对Mel频率倒谱系数特征参数进行分析研究,基于Mel频率倒谱系数改进加权函数,将体现个人语音特性的加权特征参数与反映语音帧间变化的差分Mel频率倒谱系数进行维度筛选,再进行参数混合.实验结果表明,通过改进加权函数提取得到的特征参数与差分Mel频率倒谱系数的混合参数在矢量量化的说话人识别系统中,码本容量为16和32时可以达到100%的识别率.
In the speaker recognition system,the extraction of characteristic parameters directly affects the final recognition efficiency.The study analyzed Mel frequency cepstrum coefficient characteristic parameter,improved the function of weighted Mel frequency cepstrum coefficient,and mixed the improved weighted Mel frequency cepstrum coefficient reflecting individual voice characteristics and differential Mel frequency cepstrum coefficient reflecting the voice changes between frames.The experimental results show that the mixed parameters of weighted Mel frequency cepstrum coefficient from the improved function and differential Mel frequency cepstrum coefficient makes the recognition rate of 100% when codebook-size of vector quantization is 16 or 32.
出处
《微电子学与计算机》
CSCD
北大核心
2014年第9期88-91,共4页
Microelectronics & Computer
基金
广西自然科学基金(2012GXNSFAA053221)
国家自然科学基金(61363005)
关键词
说话人识别
加权Mel频率倒谱系数
混合参数
矢量量化
speaker recognition
weighted Mel frequency cepstrum coefficient
mixed parameters
vector quantization
