Cross-spea Acoustic evidence of phonemes

74 Plot of These plots show more well-spaced than those the noun and verb aver measurements from no all four speakers are in

3.5.3 Cross-spea

In §3.5.2 I compared th speaker, looking at bot V1 position only is to e strongest and most stab in which V1=V2, and b conditioned by surroun f V1 average values of verbs only Speaker B ore clearly that the verb measurements are more se from nouns. For the individual measurement erages in this section, see Appendix C. Individ nouns are in Appendix C.1, and individual verb e in Appendix C.2. eaker vowel comparison verbs only the formant measurements of the initial stem v oth nouns and verbs. Recall that the reason for o establish a baseline of the qualities of vowel p table position. The words chosen for these meas basic infinitive verbs also ensure that these to unding vowels. In this section, I compare the vo 94 B ore symmetrical and ents which were used for idual vowel erb measurements for vowels of a single or focusing on vowels in el phonemes in their easurements i.e. nouns e tokens are not vowels of four 95 different speakers, but this time using only the initial verb stem vowels as a point of comparison, excluding nouns. This restriction is for no other reason than time limitations. The verbs used in this comparison are the same as those used for measurements in §3.5.2 above. The average formant values of all four speakers combined are in 75 below, and these figures are plotted in 76 below. The averages are based on approximately 15 tokens per vowel per speaker. 75 Average of all speakers together Verb V1 only Average F1 F2 i 341 2260 e 424 2028 498 1890 u 308 849 o 399 929 498 970 a 667 1446 76 Plot of a An interesting p above, which is related makes some observatio whether or not vowel c features. In his discussi the following Note particularl roughly to the p F1 and F2 frequ found in vowel height with no o On the other hand, whe pharyngeal cavity for a contrasting [ATR] valu f averages of all speakers together g point can be made concerning the overall shap ted to the line drawn beside the back vowels. M tions concerning the shape of the vowel plot an l contrasts are based on height features alone or ssion of a plot of Bitam Fang vowels, Maddieso larly the slope of a line connecting the back vow e position of the central vowel a … and attribu equencies co-vary in these vowels. This pattern el systems where the back series is distinguishe o other factors being significantly involved. hen another factor is involved, such as indepen r an [ATR] distinction, Maddieson suggests tha alues will have “a smaller than expected F2 diff 96 hape of the vowel plot Maddieson 2003 and its correlation with or also on tongue root son 2003 21-22 says owels which points ibutable to the fact that rn is typical of that shed by degrees of endent use of the that vowel pairs with ifference given the size 97 of the difference between the first formants” 22. For this he uses the example of Kinande, in which each ATR set shows a sloping line from high to low vowels. That line, however, cannot be drawn from vowels of one ATR set to another. Essentially, then, if Maddieson’s prediction is correct, if Ikoma’s mid-vowel contrasts are based on [ATR] and not vowel height alone, the vowel plot should not have a simple triangular shape with straight lines connecting the high, mid and low vowels. Instead, the contrasting sets, which in this case are the mid-vowel pairs, should have similar enough F2 values so that a straight line cannot be drawn to the low vowel. As 76 above shows as well as the other plots for individual speakers later in this section, this is in fact the case in Ikoma. For the back vowels especially, [o] and [ ] are quite close in F2 values at 929 Hz and 970 Hz, respectively, but F1 values show an almost 100 Hz difference at 399 Hz and 498 Hz, respectively. Though Maddieson’s claims here are not the final say in how one determines the contrastive features of a vowel system, it is interesting to at least note that Ikoma conforms to his expectations. Moving back to the cross-speaker comparison, we can further break down the vowel formant averages by comparing the averages from each individual speaker, as shown in 77 below. Speakers A, B and C are male, and D is a female. See §1.6.1 for more information about these speakers. Recall that the formant values for Speaker B were used in the previous section. 98 77 Verb V1 averages for all speakers Speaker A M Speaker B M Speaker C M Speaker D F F1 F2 F1 F2 F1 F2 F1 F2 i 351 2129 317 2323 311 2275 385 2313 e 443 1878 396 1964 396 2165 459 2103 497 1790 470 1834 502 2005 522 1930 u 296 847 289 785 280 836 367 928 o 413 898 390 897 300 948 494 974 496 968 460 958 476 913 558 1040 a 585 1316 640 1446 751 1562 693 1461 One way to view the variation between speakers is to compare the difference in the average F1 value from one vowel height to the next. This is shown in 78 below. 78 Comparison of F1 differences between speakers in Hz Speaker A Speaker B Speaker C Speaker D Height 1 to 2 u - o 101 117 20 127 i - e 79 92 85 74 Height 2 to 3 o - 70 83 176 64 e - 74 54 106 63 Height 3 to 4 - a 89 180 275 135 - a 88 170 249 171 Overall, we see that Speaker C stands out from the others in terms of vowel spacing. Strangely, there is only a 20 Hz difference in F1 between u and o. Because of this this is also a surprisingly large gap 176 Hz between o and . The difference between heights 2 and 3 and between heights 3 and 4 is also much greater than any of the other speakers, with 249 Hz and 275 Hz separating heights 3 and 4. When evaluating the other three speakers, their F1 differences are much more comparable. In all cases there is a greater difference between u – o than between i – e. Also, in all cases the change in F1 between the mid vowel mid vowels heights 1 difference between hei Finally, variatio vowel plots which show values, while the secon vowel measurements p Speaker B are in §3.5.2 79 and 80 below. 79 Plot of a els heights 2 and 3 is less than that between t 1 and 2. Note also that Speaker A has a surpris eights 3 and 4, with only 88 or 89 Hz separatin tion from one speaker to the next is most easily how two plots for each speaker. The first shows ond plot shows all individual vowel tokens. No s plotted for all four speakers can be found in A 5.2 above, so they are not repeated here. Plots f f average formant values Speaker A 99 n the high and [+ATR] prisingly small ting the two averages. ily seen in the following ws average formant Note that all individual Appendix C.2. Plots for s for Speaker A are in 80 Plot of a Speaker A has a lower vowels. Note also that Another surprising fact measurement 247 Hz be attributed to the unu Plots for Speak f all vowel tokens Speaker A er and more compressed front vowel space in c at there is an area of overlap between both sets act is that there is only a 392 Hz range from the z to the highest F1 measurement 639 Hz. Th nusually low average F1 of a, which is only 5 aker C are in 81 and 82 below. 100 comparison to his back ts of mid vowels. the lowest individual F1 This small range must 585 Hz. 81 Plot of a 82 Plot of a f average formant values Speaker C f all vowel tokens Speaker C 101 As seen above, Speake in the placement of ma some aberrant tokens o Plots of Speake 83 and 84 below. 83 Plot of a ker C has some unusual values compared to the many o tokens, which are much higher than ex s of and which are more in the e o range ker D’s average formant values and individual f average formant values Speaker D 102 the others, particularly expected. There are also ge. al vowel tokens are in 84 Plot of a Speaker D’s back mid are areas of overlap bet

3.5.4 Bandwidth