Article in The Journal of the Acoustical Society of America · December 007 doi: 10. 1121 2783198 · Source: PubMed citations 132 reads 2,169 authors
Download 358.9 Kb. Pdf ko'rish
|
IversonEvans2007
2. Vowel-space mapping
A large set of synthesized stimuli were created to map best exemplars. The stimuli were embedded in the natural carrier sentences, including the /b/ burst and the /t/ stop gap from the natural recordings. The stimuli were created using the cascade branch of a Klatt synthesizer ! Klatt and Klatt, 1990 ". For each language, the synthesis parameters were chosen so that the synthesized vowel approximated the origi- nal vowel in the natural carrier sentence in terms of F0 and amplitude contours. All other synthesis parameters were the same for each language. The upper-formant frequencies !F4=3500 Hz, F5=4500 Hz", formant bandwidths !100, 180, 250, 300, and 550 Hz for F1–F5 ", tilt !0 dB slope", and open quotient !60%" were the same for all stimuli in all languages. The stimuli primarily varied F1, F2, and duration. F3 varied as a function of F2; F3 was fixed at 2500 Hz whenever F2 was less than 2300 Hz, but otherwise F3 was raised so that it was always 200 Hz greater than F2. The F1 and F2 formant frequencies changed linearly from the beginning to the end of the vowel, and there were no additional consonantal formant transitions. F1 frequency was restricted so that it had a lower limit of 5 equal rectan- gular bandwidth !ERB" ! Glasberg and Moore, 1990 " and an upper limit of 15 ERB. F2 frequency was restricted so that it had a lower limit of 10 ERB, was always at least 1 ERB higher than F1, and had an upper limit defined by the equa- tion F2=25− !F1−5"/2. The stimuli were synthesized in ad- vance with a 1-ERB spacing of the vowel space, and with seven log-spaced levels of duration !54, 75, 104, 144, 200, 277, and 383 ms ", for a total of 109,375 individual stimuli for each language. The ERB and log-duration transforms al- lowed us to efficiently distribute the stimuli with regard to perception, although the goodness optimization procedure does not require this equal perceptual spacing. As discussed above, different natural carrier sentences had been signal processed to equate formant frequencies and F0 across the talkers of the different languages !see descrip- tion above of the natural /b/-V-/t/ recordings ", with the goal of making sure that the perceptual vowel space maps would not differ across languages due to physiological differences between the talkers’ voices !see Ladefoged and Broadbent, 1957 ". To test whether this was successful, a pilot experiment was run in which six L1 English speakers chose best exem- plars for beat, Burt, and bat in the carrier sentences for every language !subjects were asked to ignore the fact that the carrier sentences were not always in English ". The procedure was the same as for the vowel space mapping procedure described below. Repeated-measures analyses of variance !ANOVA" revealed that there was no significant effect of sentence context for any of the acoustic dimensions !F1 and F2 at the onset and offset, and duration ", p"0.05. This dem- onstrates that the acoustic normalization was successful, and that the best exemplars chosen in different languages in this experiment could thus be directly compared. Download 358.9 Kb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling