European Conference on Speech Technology
Edinburgh, Scotland, UK
The accuracy of formant frequency estimation on voiced speech in frame-based linear predictive analysis is affected by the position of the analysis frame relative to the instant of onset of vocal tract excitation. An automatic waveform-dependent point-wise analysis which employs a weighted least-square lattice (WLSL) algorithm to minimise these errors is described here. Experiments on both synthetic speech and real speech are included to show that the algorithm offers improved accuracy in comparison to the frame-based method.
Bibliographic reference. Huang, X. D. / Jack, Mervyn A. / Duncan, G. (1987): "Formant estimation based on temporal synchronous analysis", In ECST-1987, 1331-1334.