锥形核广义类时-频分布及其对语音的描述

GENERALIZED TIME - FREQUENCY DISTRIBUTION WITH CONE - SHAPED KERNEL AND ITS DESCRIBING VOICED SPEECH

语音的传统短时傅立叶分析方法不仅要假设语音具有准稳定性,而且其时间分辨率和频率分辨率存在着折衷。本文采用具有锥形核的广义类时-频分布(CK-GTFD)描述方法,并重点考察了它的瞬态响应和压缩交叉项的能力,结果表明,它不仅能同时获得好的时间分辨率和频率分辨率,对多分量信号也能精确地描述。最后,通过对语音的爆破音-元音转换段,以及元音-鼻辅音转换段的描述,表明了它在语音共振峰频谱描述、声门关闭时间确定,以及辅音-元音划分等方面的优势,为语音特征提取和识别打下了基础。

The Short - Time Fourier Analysis(STFA)of speech is based on the quasi - stationarity hypothesis, and it can't get good time and good frequency resolution simultaneously.This paper provides insights to the generalized time- frequency distribution with cone- shaped kemel(CK- GT-FD) , and develops it to the representation of important nonstationary parts of voiced speech. First, the support region of STFA, Wigner distribution and one in satisfying the weak finite time support of the generalized time - frequency distri-bution(GTFD)are investigated and plotted.Then the kernel of CK- GTFD) is derived and an algorithm is presented to maintain real transforms while utilizing an FFT of radix.2 in the transform computation. In section 3, through the responds of STFA, Wigner distribution and CK - GTFD to rapid temporal nonstationarity and two- component signal, it shows that CK- GTFD not only can provide very precise representation of time and frequency, but also has the highest suppression ability for cross- terms.Finally,the applications in plosive consonant to vowel transient and vowel to nasal consonant transient also show the great advantages of CK- GTFD in pitch period estimation, voicing detection,and formant tracking.