论文部分内容阅读
提出了一种基于多级小波分解重构和非线性动力学参数的病理嗓音基音频率检测算法。首先对病理嗓音进行多级小波分解及重构,然后采用最大李雅普诺夫指数和近似熵表征不同重构嗓音的规则度从而自适应地选择周期性最优的小波重构嗓音信号,以直接提取基音频率。实验结果表明,与传统的基音检测算法相比,该方法有效地避免了检测中所存在的倍频及分频误差,提高了病理嗓音检测的鲁棒性及准确度。
A pathological voice pitch detection algorithm based on multi-level wavelet decomposition and reconstruction and nonlinear dynamics parameters was proposed. First, the pathological voice is decomposed and reconstructed by multi-level wavelet transform. Then the largest Lyapunov exponent and approximate entropy are used to characterize the regularity of different reconstructed voice to adaptively select the periodic optimal wavelet to reconstruct the voice signal to directly extract Pitch frequency. Experimental results show that compared with the traditional pitch detection algorithm, this method can effectively avoid the frequency doubling and frequency division error in the detection and improve the robustness and accuracy of the pathological voice detection.