Effects of spectral tilt on the perception of naturally spoken vowels
James M. Hillenbrand · The Journal of the Acoustical Society of America · 2004
Experiments using synthetic signals with static spectral patterns have shown that spectral tilt can have an influence on vowel identity that is independent of the formant-frequency pattern [e.g., M. Kiefte and K. Kluender, J. Acoust. Soc. Am. 109, 2294–2295 (2001)]. The purpose of this experiment was to determine whether spectral tilt affects vowel identity for naturally spoken utterances. Test signals consisted of filtered and unfiltered versions of 300 /hVd/ utterances selected from a large, multi-talker database [Hillenbrand et al., J. Acoust. Soc. Am. 97, 3099–3111 (1995)]. Phonetically trained listeners (N=24) identified the unfiltered signals and the same signals conditioned by a +9 dB/oct filter and a −9 dB/oct filter, presented in a single random order. The effects of the two filters were clearly audible to listeners, but there was no effect of spectral tilt on vowel recognition accuracy. These findings do not invalidate earlier results with synthetic speech; rather, they suggest that when natural cues to vowel identity are preserved, overall spectral tilt plays little if any role in vowel recognition (within the limits of the ±9 dB tilt manipulation). The findings would also seem to present a significant challenge to whole-spectrum models of vowel perception, though they are by no means incompatible with this approach. [Work supported by NIH.]