基于谐波和噪声能量改进的语音截止频率轮廓估计

为了对语音谐波/噪声模型中的语音截止频率轮廓进行更好的描述,本文提出了一种基于谐波和噪声能量改进的语音截止频率轮廓估计算法。改进算法对累积谐波和噪声能量函数进行对应谐波处的功率谱加权,并且在语音截止频率轨迹的平滑部分采用形态滤波的手段。实验表明,与原算法相比,通过改进算法得到的截止频率轮廓在语谱图上标注得更为准确,在主观评分测试中改进算法也获得了优于原算法的测试评分。改进算法比原算法能够更加准确的对语音截止频率轮廓进行估计,从而使得语音谐波/噪声模型在语音编码、语音合成及识别方面具有更为有效的应用。 

Improved estimation of the voicing cut-off frequency contour based on harmonic and aperiodic energies

In order to describe the voicing cut-off frequency (VCO) contour more accurately, an improved estimation of the VCO contour based on harmonic and aperiodic energies is proposed. In the improved estimation algorithm, the cumulative harmonic and aperiodic energies function is redefined with the weighted factor of the power spectrum. And the morphology filter is applied in the VCO track smoothing section. The result shows that, the more accurate VCO contour can be labeled with the improved estimation algorithm than that with the conventional algorithm on the spectrogram. In the subjective score test, the score obtained from the improved estimation algorithm is also superior to that from the conventional algorithm. The improved algorithm can estimate more accurate VCO contour compared with the conventional algorithm to provide the speech harmonic-plus-noise model application more effectively in the speech coding, synthesis and recognition fields.