论文部分内容阅读
文章提出了一种二次谱基音检测新方法,用广义数学形态滤波与Daubechies小波相结合作为基音检测预处理,该方法是线性滤波与非线性滤波相结合的一种新的尝试。实验结果表明,该方法可以得到比传统自相关法和倒谱法更好的结果,尤其是清浊音特征不明显的音段,基音检测的性能更好,在信噪比较低的带噪语音环境下也具有很好的鲁棒性。
In this paper, a new method of pitch detection based on quadratic spectrum is proposed. The combination of generalized mathematical morphological filtering and Daubechies wavelet is used as preprocessing of pitch detection. This method is a new attempt to combine linear filtering with nonlinear filtering. The experimental results show that this method can get better results than the traditional autocorrelation and cepstrum methods, especially the voiced sound features with unambiguous voiced / unvoiced features. The performance of the pitch detection is better. In the case of noisy speech with low SNR Environment also has good robustness.