![]() It is important to have a good time resolution, hence I am interested in cross-correlation method which seems to behave better in this respect. On, at 09:36, Bogdan Rozborski via groups.io wrote: It's not free, but I think I should get it to have better view on the issue. W dniu o 16:02, Daniel Hirst via groups.io pisze: And it is important to have a good time resolution, hence I am interested in cross-correlation method which seems to behave better in this respect. I am working on an algorithm for reducing and separating noise from a periodic signal. It took me a while to internalize this idea though. I already know the 'trick' with dividing windowed signal autocorrelation function by the AC function of the window itself. I have downloaded the stuff you pointed me to. Thank you so much for such a fast and conclusive answer. Box 1642, 1000BP Amsterdam, The Netherlands Visiting address: Spuistraat 134, room 632, Amsterdam Professor of Phonetic Sciences, University of Amsterdam Paul Boersma (2009): "Should jitter be measured by peak picking or by waveform matching?" Folia Phoniatrica et Logopaedica 61: 305-308. The one in Praat is not entirely the same (it's a bit more precise, I think), but the idea is similar.Ĭross-correlation is better than autocorrelation at finding period-level variation, such as jitter and shimmer, whereas autocorrelation is better at finding intended intonation contours.įor an assessment of the cross-correlation method for jitter and shimmer, see (free from or my website): In Kleijn & Paliwal (eds.), Speech coding and synthesis. The authors did not seem to know this yet.įor a true description of cross-correlation, consult (freely downloadable from Google Scholar):ĭavid Talkin (1995): "A robust algorithm for pitch tracking (RAPT)". Boersma (1993) showed that the autocorrelation function of a signal should be correctly estimated by dividing the autocorrelation function of the windowed signal (the one that tapers off) by the autocorrelation function of the window (which "happens" to taper off in the same way). This ciriticism applies only to versions of autocorrelation before Boersma (1993). It costs 33 dollars.Īnyway, the authors seem to criticize the autocorrelation method for the same reasons that others did long ago, which is that it the autocorrelation tends to "taper off" for longer legs, due to the analysis window not being infinitely long. You can look at Samad et al 2000 for which you can find the abstract here:well, we *cannot* look at that paper, because it is not open. On, at 16:02, Daniel Hirst via groups.io wrote: What was the motivation for introducing cross-correlation method then? Could anyone point me to an article on using cross-correlation for pitch detection? Any suggestion highly appreciated.ĬNRS Laboratoire Parole et Langage (UMR 7309) - salle A304Īix-Marseille University, 5 avenue Pasteur BP 80975 As for auto-correlation method, there is Paul Boersma’s article from 1993. You can look at Samad et al 2000 for which you can find the abstract here:īefore diving into the Praat's code, I would like to learn more about methods used for pitch contour calculation.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |