Linguistics 520: Lab Assignment #5

10/2/2013 -- Due 10/16/2013

Goals:

1. Compare your vowel formants with the mean values in Hillenbrand 1995.
2. Normalize vowels as Z scores of formant values
3. Explore the "boy" and "girl" data in Hillenbrand

Put your formant measurements in a file "myvowels" that looks like this:

vowel dur F0  F1  F2  F3  F1.20 F2.20 F3.20 F1.50 F2.50 F3.50 F1.80 F2.80 F3.80
iy   153 117 252 2295 2945 273  2216  2891   251   2286 2938   266  2096  2765
ei   199 117 392 1949 2443 392  2013  2531   390   1956 2475   361  1998  2447
ae   210  99 673 1632 2316 599  1688  2328   675   1630 2314   665  1565  2358
ah   199 105 706 1199 2281 735  1280  2165   717   1219 2272   725  1354  2337
aw   204 104 613  945 2189 602   962  2195   619   1001 2209   620  1272  2184
oa   194 109 451 1005 2097 454  1008  2061   439    984 2131   409  1115  2125
uw   170 119 280  940 2125 288   989  2090   278    945 2129   283  1094  2078
ih   140 115 405 1844 2402 380  1935  2468   410   1835 2407   419  1706  2380
eh   146 106 533 1741 1924 475  1822  1984   567   1680 2269   537  1656  2307
uh   129 113 580 1265 2115 558  1305  2109   589   1296 2173   549  1427  2236
oo   120 109 416 1200 2165 387  1952  2195   422   1235 2157   412  1450  2183
er   187 111 418 1320 1578 437  1293  1608   415   1327 1613   406  1442  1680

Download the file h2.R to the same folder as your myvowels and htable files. Modify h2.R so that the line

vowelfile <- paste("YourNameHere","vowels",sep="_")

identifies you (via your "PennName").

If you execute h2.R in R, it will write out your vowel-formant data as "YourName_vowels" -- send this file to me as an email attachment.

The script h2.R will also produce the following four plots:

First, the average formant values for 12 English vowels, as pronounced by men and women in the Hillenbrand experiment, and as measured by Hillenbrand et al.:

Then a comparison of your vowels to those in the Hillenbrand data set:

Note that you should modify h2.R to compare your values to the appropriate set, so that the line that starts text(-MF2,-MF1, ... might use text(-WF2,-WF1, ... instead (and the plot title is also changed to supply your name):

MM <- read.table("myvowels",header=TRUE)
MyF1 <- MM[,"F1"]; MyF2 <- MM[,"F2"]
newyrange <- range(c(-MF1,-MyF1))
plot(-MF2,-MF1,type="n", ylim=newyrange,
     xlab="-F2", ylab="-F1", main="My data as per Praat\nCompared to Hillenbrand male speakers")
text(-MF2,-MF1,labels=vowelIDs,adj=c(0.5,0.5),col="blue")
lines(-MF2[1:7],-MF1[1:7], col="blue",lty=2)
lines(-MF2[8:12],-MF1[8:12],col="blue",lty=3)

The third graph is a scatter-plot of all F1 and F2 values for all vowels for all the Hillenbrand men and women speakers, plotted as negative F2 against negative F1 so as to imitate the orientation of the IPA vowel quadrilateral:

The fourth graph is the same thing, plotting in Z-space after converting each individual's formant values relative to his or her mean and sd of F1 and F2:

Calculate the Z-score values for your own vowel data, and create a new plot comparing your Z-scores with those in the last graph. (You may find it useful to use the "cex" argument to the points() function, in order to make your values plainly visible -- and feel free to adjust colors and so on, to the same end.)

Now, choose your own methods to explore the "boy" and "girl" data in the Hillenbrand table. One obvious thing to do is just to plot the values -- mean values by vowels, scatter plots of all vowels -- of girls against boys, or of both against women and men.

Another possibility would be to expand tables like these:

Ratio (Women/Men) of average formant values from Hillenbrand et al.

  iy ei ae ah aw oa uw ih eh uh oo er
F1 1.28 1.12 1.14 1.22 1.23 1.12 1.21 1.13 1.24 1.22 1.11 1.10
F2 1.19 1.21 1.21 1.14 1.14 1.14 1.11 1.16 1.14 1.20 1.09 1.15

Ratios across all vowels:

  F0 F1 F2 F3
W/M Ratio 1.68 1.18 1.17 1.12

And optionally, you can do some inferential statistics (is the difference between girls and boys statistically significant? What is the effect size? How do the differences compare to those between women and men?), or some statistical modeling (How do gender, age, and vowel identity interact? What about individual factors within gender and age groups?)