We have defined and evaluated a set of vowel and non-vowel sounds that were matched for acoustical features but which differed markedly in their similarity to speech [1]. Four damped sinusoids [2] with formant frequencies were used to produce artificial vowels; they were repeated with a period in the vocal pitch range, and had half lives that produced proper formant bandwidths. These artificial vowels automatically activate the phonological system, producing a clear speech perception (listen to dv). Non-speech control sounds with similar long-term distributions of energy over frequency and time are generated by randomising the envelope onsets and the carrier frequencies (listen mr) within each of four formant tracks. Four classes of these sounds are shown in Figure 1: speech-like with periodicity pitch or no pitch, and non-speech like with periodicity pitch or no pitch. Figure 2 shows the respective spectrograms for one example of each of these stimulus classes.

Figure 1: Four classes of stimuli constructed from sets of isolated formants (damped sinusoids). Top left: "damped" vowel, dv, with regular envelope onsets and fixed formant frequencies. Top right: "raspy" vowel, rv, with fixed formants but randomised envelope onsets. Bottom left: damped sinusoids with synchronous onsets (which produces a pitch), but carrier frequencies which are randomised from cycle to cycle. It sounds like a gurgle with a pitch ("musical rain with pitch", mp) Bottom right: both onsets and carrier frequencies are randomised from cycle to cycle, producing "musical rain", mr.

Figure 2: Spectrograms of the same stimuli as in Figure 1: top left: dv; top right: rv; bottom left: mp; bottom right: mr.

References

[1] Uppenkamp S, I Johnsrude, D Norris, W Marslen-Wilson, RD Patterson (2006) Locating the initial stages of speech-sound processing in human temporal cortex. Neuroimage, accepted for publication 16 January 2006.
[2] Patterson RD (1994) The sound of a sinusoid: time-interval models. J Acoust Soc Am 96, 1419-1428.