 |
|
|
|
|
| 1 | 2 | 3 | 4 | 5 |
(see AUDIO FILES for information on the *.wav files)
In 1996 Yost, Patterson, and Sheft (see references below) showed that it was highly unlikely that discriminations involving iterated ripple noise stimuli could be based on spectral processing. Their analysis suggested that temporal processing, as might be realized by an autocorrelation type mechanism, could describe the iterated ripple noise discriminations. Figure 1 shows the two basic iterated ripple noise networks and Figure 2 shows the three basic discriminations that were fundamental in leading Yost et al to their conclusions. The discriminations were between an IRNO stimulus with 2 iterations and IRN stimulus with 1 iteration (top panel of Figure 2), between IRNS stimulus with 2 iterations and IRN stimulus with 1 iteration (middle panel), and between IRNO and IRNS stimuli each with 2 iterations (bottom panel). As can be seen the spectral differences between the two stimuli in the top and bottom panel are very similar and there are several spectral differences that could be used to aid discrimination (e.g., the difference in the width of the spectral peaks and the difference in the smaller peaks in the valleys between the major peaks). The spectral differences in the middle panel are less obvious.
For each discrimination described above there are five examples of pairs of two stimuli. For each pair you are to decide if there is a difference between the two stimuli or if the two are the same. That is, this is a five-trial, same-different experiment. The Answer for each discrimination will indicate the "correct" answers.
IRNO(d=4ms,g=1,n=2) versus IRN(d=4ms,g=1,n=1)
|
|
|
|
|
| 1 | 2 | 3 | 4 | 5 |
IRNS(d=4ms,g=1,n=2) versus IRN(d=4ms,g=1,n=1)
|
|
|
|
|
| 1 | 2 | 3 | 4 | 5 |
IRNO(d=4ms,g=1,n=2) versus IRNS(d=4ms,g=1,n=2)
|
|
|
|
|
| 1 | 2 | 3 | 4 | 5 |
Most likely you were correct for the first two sets of discriminations and were unable to make correct discriminations for the last set of discriminations. That is, it is easy to tell IRNO(4,1,2) from IRN(4,1,1) and IRNS(4,1,2) from IRN(4,1,1), but very difficult to tell IRNO(4,1,2) from IRNS(4,1,2). This does not appear consistent with the spectral differences shown in Figure 2.
The autocorrelation functions for the same comparisons that were used for Figure 2 are shown in Figure 3. Note the first peak (at 4 ms) in the autocorrelation functions. Its height is 0.67 for the IRNO(4,1,2) and IRNS(4,1,2) stimuli and 0.5 for the IRN(4,1,1) stimulus. Thus, if the height of the first peak in the autocorrelation function were the basis for predicting discrimination performance, then IRNO(4,1,2) should be discriminable from IRN(4,1,1) and IRNS(4,1,2) should be discriminable from IRN(4,1,1) [0.67 is different from 0.5]. But, since both IRNO(4,1,2) and IRNS(4,1,2) have the same autocorrelation function first peak height [0.67], they should be difficult to discriminate. These are the typical results. Thus, analysis based on autocorrelation is consistent with the discrimination results, whereas analysis based on spectral comparisons is not. Additional discussion of the use of autocorrelation in ripple noise pitch processing can be found in the references given below.
High-Passed Filtered IRN stimuli
Another way we (see Patterson et al, 1996) have tried to
demonstrate that IRN processing is most likely temporally based
is to high-pass filter IRN stimuli such that the spectral ripples
would not be resolved by the auditory system. Consider an
IRNO(16,1,8) stimulus. This stimulus has its spectral peaks
spaced every 62.5 Hz (1/16 ms). If we high-pass filter this above
1000 Hz, it is unlikely that the auditory system can resolve this
62.5 Hz spectral ripple. Figure
4 shows the neural activation pattern (the NAP) for a
high-pass filtered (1000 to 4000 Hz) version of IRN(16,1,8). The
NAP (see Patterson et al, 1995) is generated by first using a 64
channel gammatone filter bank to simulate the mechanical tuning
of the basilar membrane. The outputs of each filter are then
rectified, power-law compressed, and low-passed filtered at 1200
Hz to simulate the neural response to the basilar membrane
pattern. The resulting NAP shows the neural output for each
channel (channel number is not to scale, but the center frequency
in Hertz of the lowest and highest channel are given on the right
of Figure 4). That is, each line in the pattern represents the
neural activity in a fiber or group of fibers tuned to a narrow
frequency region. As one can see it is difficult todetermine any
particular pattern to the NAP. Patterson et al (1996) show that
it is difficult, if not impossible, to tell this NAP from one
generated with a noise that is not rippled. However, as Figure 5
shows there is a
temporal structure to the information available in the NAP. If
the output of each channel is autocorrelated then the
"correlogram" of Figure 5 reveals a strong correlation
of "neural" activity at 16 ms, the delay used to
generate the ripple noise. It is relatively easy to hear the
difference between a 1000-Hz high-passed filtered noise and a
1000-Hz high- passed filtered IRNO(16,1,8) stimulus as the
audio example shows. So in the absence of resolvable spectral
peaks, but with strong temporal correlation at the delay of 16
ms, it is possible to perceive the temporal regularity of the IRN
stimulus.
Suggested References
Yost, William A., Models of the Pitch and Pitch Strength of Ripple Noise, Journal of the Acoustical Society of America, 66, 400-411, 1979
Yost, William A., The Dominance Region for the Pitch of Ripple Noise, Journal of the Acoustical Society of America, 72, 416-426, 1982
Yost, William A., Patterson, R.D., and Sheft, S. A Time Domain Description for the Pitch Strength of Iterated Ripple Noise, Journal of the Acoustical Society of America 99, 1066-1078, 1996
Patterson, R.D., Allerhand, M. H., and Giguere, C. Time Domain Modeling of Peripheral Auditory Processing: A Modular Architecture and a Software Platform, Journal of the Acoustical Society of America, 98, 1890-1894, 1995
Patterson, R.D., Handel, S., Yost, William A., Datta, J., The Relative Strength of Tone and Noise Components in Iterated Rippled Noise, Journal of the Acoustical Society of America, 100, 3286-3294, 1996
Yost, William A., Pitch Strength of Iterated Rippled Noise, Journal of the Acoustical Society of America, Journal of the Acoustical Society of America 100, 511-518, 1996
Yost, William A., Pitch of Iterated Rippled Noise, Journal of the Acoustical Society of America, 100, 511-518, 1996
