Figure 3.

Auditory sensitivity for the direction of periodic sounds in the frequency domain. Shepard sequences [69] were created as a superposition of pure tone sweeps, increasing or decreasing in frequency over time. The sweep speed varied for different sequences: each tone increased or decreased logarithmically between the boundaries of audible space (set between 80 and 11025 Hz) over a fixed duration T; a new tone was inserted into the sequence (and an old tone disappeared) every T/40, such that at every instant 40 sweeping tones were simultaneously present. The temporal frequency of this periodic sound motion, therefore, was defined as TF = 40/T. In different trials, this temporal frequency was varied between 1 and 8 Hz. Each sequence was 12 s long, with a Gaussian amplitude profile in both time and frequency space, to limit the perception of artefacts linked to sequence onset/offset and tone insertion/disappearance, respectively. Examples of upward and downward sequences for TF = 1 Hz and TF = 8 Hz are presented as spectrograms in the figure (colour map indicates stimulus energy at each time and frequency), and the corresponding sound files can be downloaded from http://www.cerco.ups-tlse.fr/~rufin/audiovisual/. Participants (n = 6) were presented with 10 sequences of each temporal frequency in randomized order, moving up or down in frequency space (randomly determined with 50% probability), and asked to report their perceived motion direction (up/down) by pressing arrow keys on the keyboard. To limit the possibility of relying on perceived pitch differences between upward and downward sequences, the frequency-domain Gaussian amplitude envelope (s.d. 0.25 log units) was centred at one of three frequencies (700, 800 or 900 Hz), randomly chosen for each trial. The direction judgements were only accurate up to 3–4 Hz (t-test against 0.5 = chance-level, p < 0.05), and deteriorated rapidly at higher temporal frequencies. This low-pass sensitivity function critically limits the possibility of measuring a c-WWI effect in the auditory domain. Error bars represent standard error of the mean across subjects. (Online version in colour.)