论文部分内容阅读
本文介绍一种预分浊音型的LPC基音提取算法,对语言信号先用线性预测系数a_0和a_1的差值分出浊音区,然后只对浊音部分进行基音提取。提取基频时,数据率减半,用LPC的自相关方法产生8个预测系数的倒滤波器,倒滤波后的误差信号,用平均幅差函数(AMDF)方法提取基频,再线性插值,最后用非线性平滑滤波,并将所得结果和一个半自动的精确算法,以及简化倒滤波(SIFT)算法进行比较,说明我们提出的算法,对背景噪声40dB以下的连续语言是准确有效的。它避免了清音和无声间隙区的音调计算,且浊音和清音的判别比较准确。
In this paper, we introduce a preprocessing LPC pitch extraction algorithm, which separates the voiced region from the difference between the linear prediction coefficients a_0 and a_1 for the speech signal, and then only performs pitch extraction on the voiced part. When the fundamental frequency is extracted, the data rate is halved. The LPC autocorrelation method is used to generate an inverted filter of eight prediction coefficients. The filtered error signal is filtered and the fundamental frequency is extracted by the method of average amplitude difference (AMDF) Finally, the nonlinear smoothed filter is compared with a semi-automatic accurate algorithm and a SIFT algorithm to show that the proposed algorithm is accurate and effective for continuous speech with a background noise less than 40 dB. It avoids the tone calculation of unvoiced and silent gaps, and the discrimination between voiced and unvoiced sounds is more accurate.