本文对确定性窄带信号的理论相关函数和信号样本以及随机窄带信号样本用自回归(AR)和FFT两种方法进行了谱估计。这些结果很好地解释了声达信号的谱估计结果,即对靠得很近的双谱峰,AR的分辨率劣于FFT,对于相距较远的双谱峰和单谱峰,AR的估计精度好于FFT。在所有的情况下.AR方法的谱曲线平滑,旁瓣低。这两种方法联合使用是对声达信号进行谱估计的好方法。 In this paper, the theoretical correlation function and signal samples of deterministic narrowband signal and the random narrowband signal samples are estimated by autoregressive (AR) and FFT methods. These results provide a good explanation for the spectral estimation of the vocal signals, ie, the resolution of the AR is inferior to that of the fast-approaching bispectrum and the estimation of the AR for the bispectrum and singlet far apart Accuracy better than FFT. In all cases, the spectral curve of the AR method is smooth with low sidelobes. The combination of these two methods is a good way to estimate the spectrum of the acoustic signal.