摘要
针对语音产生的数字模型来分析清音和浊音的特点,研究了清音和浊音的压缩感知观测序列的特性,从而推出清音的压缩感知观测序列具有近似高斯信号的特性,而浊音的压缩感知观测序列具有非高斯信号的特性。基于这种特性来设计一种直接针对压缩感知观测序列、非重构的清浊音判别方法,并给出了它与重构信号清浊音判别在准确性和计算量两个方面的比较结果,解决了基于语音压缩感知观测序列重构情况下判别清浊音的高计算量问题。
Based on the theory of compressed sensing, the observation sequence after compress- ing sensing is different from the Nyquist sequence, so the voicing-state identification can be a- chieved only by reconstructing the original speech signal with high complexity. The voicing- state characteristics are analyzed based on the speech digital model, and a conclusion can be drawn that the unvoiced observation sequence has the characteristics of Gaussian signal while the voiced observation sequence has the characteristics of non-Gaussian signal. According to the characteristic, a voicing-state identification algorithm of third-order accumulation is de- signed based on observation sequence, and is compared with the energy discrimination method of the reconstructing speech signal in accuracy and computing. Therefore, the problem of high complexity in voicing-state identification can be solved after reconstructing the original speech signal.
出处
《数据采集与处理》
CSCD
北大核心
2013年第3期274-279,共6页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(60971129)资助项目
关键词
压缩感知
清浊音判别算法
观测序列
三阶累积量
compressed sensing
voicing discrimination algorithm
observation sequence
third-order accumulation