Notes from Friday, July 27: 1) Try to extract for just one speaker from the Hillenbrand et al. corpus, a set of values to plot in something like Figure 1 in the article. (Don't spend more than 20 minutes on this.) What this means is to get a set of numbers for the first four formants into a plain text file that we can read as a table into R. Ideally, these numbers will be N sets of four values, where N is the length of the vowel divided by 0.008 (i.e., the 8 ms steps that are described for the "16 ms hamming windowed segments" on p. 3100. No one was able to do this successfully in the 20 ms allocated to it. We discussed how to do this, but David also pointed out that we do want to have some data reduction in order to relate our values to the plots in this paper (and in other papers that he is grappling with in the sociophonetics literature). We wondered about how to extract numbers to correspond to the 8 points per vowel in the bigdata.dat file from the web page for the paper. Mary will make a script to do either this (or some other sampling that we decide on) over the weekend. In the meantime, we should decide what grain is most relevant for going beyond these data to the less controlled recordings that Jocelyn and David want to analysis. ==> David wants to do 20% and 80% ... 2) Play with the R script that you can download from http://ling.osu.edu/~mbeckman/795.10/scripts until you feel comfortable with it. Try to figure out what some of the commands do. We resolved some of the problems with path names and so on that several of us had had in trying to run the script on their own laptops. We played with modifying the last loop in this file to superimpose the women's values and the kids' values onto the men's values in a plot relevant to what David wants to do for 1).... 3) Think more about how to represent "the phonetic space" Indirectly related to this goal, we looked at Bark space "spectrograms" of the Korean words that Mira showed us last time illustrating /twi/ vs /ti/ and /kwi/ vs /ki/, in order to appreciate the point that our resolution of frequency differences in the region of a /t/ burst is much less good than our resolution of differences in the region of a /k/ burst, so that even if the spectral effects on the stop release of the concurrent labiovelar gesture in /twi/ and /kwi/ were an identical lowering of the dominant peaks in the spectrum, these physically identical effects would have different perceptual consequences. This is relevant for evaluating distances in the F1-F2 space. For example, a 200 Hz difference for F2 in the /u/ vs /U/ F2 region will not be perceptually equivalent to a 200 Hz difference in the /i/ vs /I/ region. ======================================================== Preparation/schedule for classes next week. We decided that we would break up next week for individual meetings for Tuesday through Thursday and will meet in Mary's office for these. Jeonghwa and Mary will start working on prosodic perception on Tuesday. We'll start working with a set of recordings for Egnlish and for Korean. David and Mary will meet on Wednesday. David will prepare by transcribing (in plain text) one of the recordings he wants to analyze. We will work on translating the transcript into a useful TextGrid file. Jocelyn is free to attend this session if she wants. Jocelyn and Mary will on Thursday. Jocelyn will TextGrid tag all of the sentences for all six talkers. We will then start working on how to use the TextGrids for transcription of the key words and the acoustic analysis of the vowels in the key words. We will then meet again on Friday in 29 Derby to assess what we've done and go on from where we ended in the individual sessions. From: ddurian@ling.ohio-state.edu Subject: Northern California vs. Northern Cities Vowels Date: Fri, 04 Aug 2006 15:33:54 -0400 Hi all, Since this came up today while talking about /twi/ vs. /ti/--here's an interesting comparison of Jocelyn's home dialect vowel system and my (David's) home dialect system. http://www.stanford.edu/~eckert/vowels.html David