Fig 1.
Characterization of a palatal click.
The representations are respectively a temporal (a) and a spectral (b) plot of the tongue click used in this study. The spectrum was computed by Fourier transform without windowing.
Fig 2.
Continuous wavelet transform of a tongue click.
The duration of the wavelets was set to 7 samples at 44.1kHz, i.e. 159μs. The spectro-temporal representation was computed up to 12kHz and was normalized in magnitude.
Fig 3.
Drawings of the scenarios with nearby walls with different textures.
The respective textures are a flat wall (a) a circular wall (b), a wall with aperture (c), a parabolic wall (d), a crenelated wall (e) and a staircase (f). The source is labelled as S. Dimensions are expressed in cm.
Fig 4.
Drawings of the scenarios with far-away walls (5.0m) with different textures.
The respective textures are a flat wall (a) a circular wall (b), a wall with aperture (c), a parabolic wall (d), a crenelated wall (e) and a staircase (f). The source is labelled as S. Dimensions are expressed in cm.
Fig 5.
Normalized impulse responses of nearby walls (0.8m) with the following textures.
Fig 6.
Spectral difference between the direct and reflected components (unconvolved) for each texture at 0.8 m distance.
The spectra were obtained by Fourier transforming the two components without windowing, starting 0.05ms before the maximum of the direct sound and 3.85ms later for the reflected sound. The black and light grey curves respectively correspond to the direct and reflected components of the different binaural impulse responses. The spectra were normalized at 1.5kHz for the sake of better visualizing the coloration difference.
Fig 7.
Normalized impulse responses of far-away walls (5.0m) with the following textures.
Fig 8.
Spectral difference between the direct and reflected components (unconvolved) for each texture at 5.0 m distance.
The spectra were obtained as described in the caption of Fig 6 except for the reflected sound, starting 20.6ms after the peak of the direct sound. The black and light grey curves respectively correspond to the direct and reflected components of the different binaural impulse responses. The spectra were normalized at 1.5kHz for the sake of better visualizing the coloration difference.
Fig 9.
Texture discrimination performance grouped by distance as tested on 14 sighted participants.
The level of accuracy corresponding to the guessing limit (50%) is represented by a dark black dashed line.
Fig 10.
Texture discrimination performance for the three configurations of impulse responses tested on 14 sighted participants.
The level of accuracy corresponding to the guessing limit (50%) is represented with a dark-black dashed line.
Fig 11.
Texture discrimination performance for the interaction between the three configurations of room impulse response and the 2 distance groups investigated, tested on 14 sighted participants.
The level of accuracy corresponding to the guessing limit (50%) is represented with a dark-black dashed line.
Fig 12.
Overall texture discrimination performance tested on 14 sighted participants.
The level of accuracy corresponding to the guessing limit (50%) is represented with a dark-black dashed line.
Fig 13.
Audibility performance of the stated textures per distance group tested on 14 sighted participants.
The level of accuracy corresponding to the guessing limit (50%) is represented with a dark-black dashed line.
Fig 14.
Listening test based statistical significance (p-value) of the discrimination fraction of the presented texture pairs at the two investigated distance.
The right and left sections of the table respectively correspond to 0.8m distance and 5.0m distance. The pairs that are significantly discriminated are greyed out. The included drawings provide a schematic representation of the textures and depict neither the size nor the layout of the simulated ones.
Fig 15.
Texture discrimination performance per configuration of room impulse response tested on 14 sighted participants per distance group.
This figure is split into 3 parts, respectively corresponding to the nearby walls (a.), the far away walls (b.) and textures at both distances (c.). The level of accuracy corresponding to the guessing limit (50%) is represented with a dark-black dashed line.
Fig 16.
Correlation between the difference between time signals of texture pairs (SNQ1) and discrimination performance.
SNQ1 consists in the rms value of the difference between the time signals of each pair of textures. The dashed light grey curves are 3rd order polynomial fits of the rms difference between the time signals with the related discrimination factors.
Fig 17.
Correlation between the difference between spectra of texture pairs (SNQ2) and discrimination performance.
SNQ2 consists in the rms value of the difference between the frequency domain of each pair of textures. The dashed light grey curves are 3rd order polynomial fits of the rms difference between the time signals with the related discrimination factors.
Fig 18.
Correlation between the difference of non-smoothness spectra of texture pairs (SNQ3) and discrimination performance.
SNQ3 consists in the absolute value of the difference between the rms values of the numerically differentiated spectra of each pair of textures. The dashed light grey curves are 3rd order polynomial fits of the rms difference between the time signals with the related discrimination factors.
Fig 19.
Correlation between the ANN-predicted discrimination performance and the discrimination performance that came out from the listening tests.
The dashed light grey curves are 3rd order polynomial fits of the rms difference between the time signals with the related discrimination factors. Plus symbols are for the 15 training data, circles are for the 15 test data.