I have sibilant fricative productions- and I want to "get rid" of the anatomical variation due to sex differences. Any formula that the community knows of which has proven to be reliable. Or a good literature recommendation?
Dear Xaver Koch!
I can recommend you Discriminant Analysis as a general approach. You must find characteristics of your data that are not sex-dependent.
Hi Joseph,
OK, thanks for the recommendation. However for vowel formant values there are rough estimations on how much pure anatomical changes (length of the vocal tract above the glottis) affect on acoustics. Just thought that maybe formulae would be out here in the field.
I think this is not a trivial problem, because the vowel space in females is not only just bigger than in males, since vowels are affected differently (e.g. F1 differences in /a/ are huge, but not in /i/; see work by Melanie Weirich who is also in research gate who works a lot on male-female differences in acoustics and articulation; some of the relations/formula have been describd in the work by Gunnar Fant, sorry that I do not know the exact reference now). I think that there are several methods for normalization. Maybe it is a matter of what you want to do further with the data (rather technological use, or should this be very precise or for statistics maybe even z-transformation may help...).
Hi Xaver, do you have enough speakers per gender group to do a within-gender z-transform (as Susanne suggested) on your measures of interest? That seems like a defendable general approach to me.
Further, I don't know of any formula on gender differences in fricatives, but I think the person who would know if that exists would be Allard Jongman (University of Kansas).
Many researchers have attempted to normalize spectra to account for variability in speech between male and female speakers in the hope that they can attain invariance. Also, investigators have attempted normalization of vocal tracts size for the same purpose. But consider the implications: If you normalize the spectra, you will obtain a signal which was not produced by a male or female speaker. In fact the normalized signal was not produced by anyone and so is not truly a valid stimulus for a theory of hearing. Efforts are geared toward biomechanical invariance in order to see the way the sonorous body transforms sound wave. This is not a simple problem that can be resolved from the acoustic signal in the absence of invariance. Please try any method you can find, but you will get to know that it is as yet impossible. For some deeper insights, you might want to look at the volune PItch and Neural Coding; Ed Plack and Oxenham (2005) Springer; Ecological Psychoacoustics Ed Neuoff (2004); Hardcastle, Laver and Gibbon, Eds Handbook pf Phonetics Sciences, (2010). You can also read my paper The tension theory of pitch production and perception (2014) Don't lose courage.
Sorry, I did not read carefully enough that you have sibilants. Most of the literature on differences between males and females in /s/ production consider these differences as intended (to signal gender) - not anatomical. The argument is that at the place where sibilants are realized (at the alveolars) is a place where no sex specific differences in anatomy of the vocal tract are found (they are more in the pharynx - but what was never investigated is the length of the incisors or things like that). If you find differences in your data, they might have a social origin (see also work by Benjamin Munson on homosexuals or Jane Stuart Smith). That doesn't help now for normalization.
Thank you all for you thoughts and ideas. Nice thing is that I have both sibilant and vowel values for my (>100) speakers (60%f : 40%m). Infact I see sign. COG differences - which I interpreted - in line with Susanne - as intended (gender signal). Just wanted to make sure that the normalization issue, which is quite common when discussiing vowel results doesnt pop up for my sibilants. Maybe I should nevertheless also analyze the vowels and display gender differences for consonants in comparison to gnder differences for vowel spaces and contrasts. Best X.
Hello, on this web page http://ncslaap.lib.ncsu.edu/tools/norm/norm1.php you can choose some of normalization methods
Looks like the old link does not work anymore, here is the working one:)
http://lingtools.uoregon.edu/norm/about_norm1.php
@Xaver Koch Hey, I am wondering about a similar issue. I have COG data of l, t, n, s from male and female speakers. When I run a statistical analysis, it shows that the gender differences outweigh any other effects. I think I should normalize my data to get another interpretation of my data. I know It has been a while now since you started this thread but how did it go for you?
- Badges
- Science topic
- Similar topics
- Geoscience
- Cartography
- Mapping