一种适合于宽带语音的清浊音分类算法

Voiced/Unvoiced Classification Algorithm for Wideband Speech

提出了一种在宽带语音(50—7 500 Hz)中进行清浊音分类的方法,该方法通过估计和跟踪信号的高频能量和低频能量提取语音信号,并针对宽带语音的特点,用特殊的高低频谱能量比和周期性水平组合参数进行清浊音分类。算法设计了一种可以在所有语音帧更新阈值的方法,在比较复杂的背景噪声条件下仍能较准确地更新阈值,从而保证清浊音分类有效进行。仿真结果表明,本文方法能准确地对宽带语音进行分类,具有一定的抗噪声性能,且算法的复杂度较低,具有一定的实用价值。

A voiced/unvoiced classification algorithm is presented based on wideband speech （50--7 500 Hz）. The algorithm extracts the speech signal by tracking the high and low frequency energyies of the noised signal. Then a combination with the periodicity level and special high-to-low frequency spectrum energy ratio is used to classify the voiced/unvoiced speech frames according to the characteristic of the wideband speech. A method is designed by the algorithm for updating the threshold value, and it can update the threshold under complex background noise. The method can ensure the voiced/unvoiced classification algorithm to work efficiently. Experimental results show that the algorithm has high voiced/unvoiced classification accuracy and comparatively low complexity.