https://wenku.baidu.com/view/f2ce376ea98271fe910ef9ce.html
https://github.com/larrymario/LipSyncLite
The concept of
formant Formant is an important feature that reflects the resonance characteristics of the vocal tract. Humans use formant information in speech perception. The formant information is contained in the frequency envelope. The key to extracting formant parameters isEstimate natural speech spectral envelope,Generally consideredThe maximum value in the spectrum envelope is the resonance peak.
Using the corresponding low-frequency part of the speech spectrum Fourier transform to perform inverse transformation, the envelope curve of the speech spectrum can be obtained.
The first fourth resonance peak is determined according to the magnitude of the peak energy of the spectrum envelope.
Method of extracting formants The method of extracting formants
based on linear prediction (LPC) is for further study.
There are two ways to choose linear prediction formants: one way is to use a standard program to find complex roots to calculate the roots of the prediction error filter, calledRoot finding; Another way is to find the local maximum in the spectrum envelope derived from the prediction, calledPeak selection。
Cepstrum method:
The cepstrum of the channel response decays very quickly, and the value outside [-25,25] is very small, so a corresponding cepstrum filter can be constructed to separate the cepstrum of the channel, and the pairs are separated Do the corresponding inverse transformation of the cepstrum to obtain the logarithmic spectrum of the vocal tract function. Further processing can be performed to obtain the required resonance peaks.