Estimation of the formant frequencies by means of a wavelet transform of the speech spectrum
Mounir Bensaid, Jean Schoentgen · 1997
The objective is to present a method that extracts the spectral envelope of a speech signal. The method is based on the wavelet transform, which is a family of multi-resolution analysis methods. The extraction is founded on the observation that spectral envelope and source-related spectral components vary on different frequency scales. The difference between cepstral and wavelet analysis is that the latter is not based on the hypothesis that a speech signal is the outcome of a convolution of the source signal with the vocal tract impulse response. The method of analysis was tested on a corpus of [a],[i],[u] vowels sustained by healthy and dysphonic speakers. Results show that the envelopes extracted via cepstral and wavelet analysis are very similar except in the case of [u], for which the first two formants extracted by means of wavelet analysis are shifted slightly towards higher values and the formant peaks are somewhat closer. Keywords--- Signal Analysis, Cepstrum, Wavelet transfo...