| 摘要 |
A method of detecting a formant frequency of a voice signal is provided to detect the formant frequency delicately without a large amount of operation by polishing roots at the maximum value that has overlapped formants. A voice signal is filtered(SP22) and then a proper window is multiplied to the voice signal so as to extract a short time signal(SP23). A linear prediction coefficient is calculated by the short time signal and then a linear prediction coefficient error filter is obtained with the linear prediction coefficient(SP24). A linear prediction spectrum is obtained by using the linear prediction coefficient error filter(SP25) and then a formants frequency candidate is extracted from the maximum values of the linear prediction spectrum(SP26). It is determined whether the number of poles is larger than one when the formants frequency candidate is seen at a z-area with a Cauchy's Integral Formula(SP27,SP28). If the number of the poles corresponding to an area where a peak exists at a spectrum of the z-area is larger than one, roots are polished so as to be extracted(SP29) and a formants frequency is obtained by use of the extracted roots(SP30).
|
