Detection of stop consonants with the wavelet transform

We present the burst detection of stop consonants in the French language with the wavelet transform. The system is based on the study of normalized correlation functions between the modulus of the impulse wavelet transform and the same transform of the speech signal. Every correlation function presents a minimum before each maximum, where that maximum is synchronous with the burst. Two signals S/sub M/ and S/sub m/ are computed. They respectively are the synchronous and asynchronous sum of all correlation functions. The detector presents two parts: first the localisation of a local minimum of the S/sub m/ function is found, then the temporal localisation of the burst is accomplished with the next maximum of the S/sub M/ function. The method presents a best compromise between the number of omissions and insertions for the unvoiced stop consonants. The main interest of this method is the good time localisation of the discontinuity. Furthermore, it shows that for detection, the burst can be modelled with a pulse.<
>