Spectral/temporal trade-off

Zatorre, R.J. & Belin, P. (2001) Spectral and temporal processing in human auditory cortex.  Cerebral Cortex, 11, 946-953.

We used PET to examine the response of human auditory cortex to spectral and temporal variation. Volunteers listened to sequences derived from a standard stimulus, consisting of two pure tones separated by one octave alternating with a random duty cycle. In one series of five scans, spectral information (tone spacing) remained constant while speed of alternation was doubled at each level. In another five scans, speed was kept constant while the number of tones sampled within the octave was doubled at each level, resulting in increasingly fine frequency differences.  Results indicated that (1) core auditory cortex in both hemispheres responded to temporal variation, while anterior superior temporal areas bilaterally responded to the spectral variation; and (2) responses to the temporal features were weighted towards the left, while responses to the spectral features were weighted towards the right. These findings confirm the specialization of left-hemisphere auditory cortex for rapid temporal processing, and indicate that core areas are especially involved in these processes. The results also indicate a complementary hemispheric specialization in right-hemisphere belt cortical areas for spectral processing. The data provide a unifying framework to explain hemispheric asymmetries in processing speech and tonal patterns. We propose that differences exist in the temporal and spectral resolution of corresponding fields in the two hemispheres, and that they may be related to anatomical hemispheric asymmetries in myelination and spacing of cortical columns.


The stimuli consisted of pure-tone patterns in which the frequency and duration of the individual elements were varied systematically. At each frequency transition, tones were ramped on and off in counterphase using a cosine function with a 10 msec duration, thus maintaining a nearly constant total amplitude envelope and avoiding transients. Two stimulus parameters, rate of frequency modulation, and spectral distribution of elements within the pattern, were varied independently in the two conditions of the experiment, while keeping the frequency range and amplitude constant at all times.

The parametric changes in the two dimensions were applied starting with the same standard stimulus condition (ST), consisting of two tones separated by one octave with a relatively slow rate of alternation. This standard condition was then subjected to five levels of temporal variation and five levels of spectral variation, for a total of 11 scans. Throughout a given scan the parameters were held constant, so that variation in the desired parameter occurred only across scans.

In the temporal condition, the frequencies used were fixed at 500 and 1000 Hz, thus keeping a constant separation of 1 octave. The rate of alternation of these two frequencies was varied from relatively slower to relatively faster across each of five different scans, doubling the speed with each successive condition. The value of the parameter t varied from the slowest, at 667 msec (ST), to the next faster with t=333 msec (T1), and so forth to the fastest condition (T5), where t=21 msec.

In the spectral condition, a sequence of tones sampled from between 500 and 1000 Hz, was presented according to the same random duty-cycle parameters as described above, using a fixed value of t=667 msec. The frequency differences between tones were varied parametrically from coarsest to finest within each of the 5 scans, using frequency separations of half an octave (S1), one-quarter octave (S2), one-eighth octave (S3), one sixteenth octave (S4) and 1/32th octave (S5). Thus, starting with the standard condition (ST), the stimuli consist simply of the same two tones one octave apart as described above. The change across conditions is thus related to the fact that the number of different frequencies sampled within any given frequency band, integrated over some time window, increases as a function of the spectral parameter.

Note that in the spectral conditions, only the distribution of spectral elements (i.e., number of frequencies) varies across scans, with no change in rate of presentation, whereas in the temporal conditions the converse is true: only the rate is varied and the spectral difference is kept constant at one octave.  Click on each image below for a short sample of the stimulus.


667 ms 1 octave

333 ms

1/2 octave

166 ms

1/4 octave

83 ms

1/8 octave

41 ms

1/16 octave

21 ms

1/32 octave

Note: All sounds on this page are MP3 files encoded at 64 Kbps and are of lower audio quality than the stimuli used in the actual experiments.  If your browser is not configured to launch an MP3 player automatically, right click on the link and choose "Save link as..." in Netscape Navigator or "Save Target As..." in Microsoft Internet Explorer to save the file to disk.  You can then play the sample with any MP3 capable media player such as the one below.  It has recently been verified that, for reasons still unknown, some versions of Netscape Navigator 4.x can sometimes save MP3 data incorrectly.  Mozilla, Opera, Netscape Navigator 6.x and Internet Explorer do not have this difficulty.