Abstract:
The aim of this work was to determine the effect of noise presence at time to evaluate the pitch, and the study of possible improvements. Several algorithms (Terhardt, Duifhuis, Cepstrum, simple FFT, and visual determination) were studied. These algorithms were applied to different Spanish vowels with a progressively increased noise level; conclusions about the precision of each method are presented. In general, the noise causes a loss of accuracy because time-domain data are blurred and spurious components appear in the spectrum. At this point the multiresolution analysis (MRA) technique becomes a useful tool to eliminate the effect of elevated noise level. MRA was used to smooth the vowels analyzed and then the algorithms under study were applied another time. With this procedure an accuracy improvement in all methods that work in the frequency domain was observed.