A cepstral domain algorithm for formant frequency estimation from noise-corrupted speech

A new scheme for the estimation of formant frequencies from noise-corrupted speech signals is presented in this paper. In order to overcome the effect of noise, first, instead of conventional autocorrelation function (ACF), a once-repeated ACF of the observed data is employed. A ramp cosine cepstrum model of the ORACF of speech signal is developed, followed by a model-fitting based least-square optimization to extract the formants. For the purpose of implementation, the discrete cosine transform (DCT) is used which offers computational advantages for real signals and solves the phase unwrapping problem. Synthetic and natural vowels as well as some naturally spoken sentences in noisy environments are tested. The experimental results demonstrate a better performance obtained by the proposed scheme in comparison to some of the existing methods at low levels of signal-to-noise ratio.