Note: if you are using IE 6 with a screen size of 800X600 or smaller and would like to see the enlarged image at full resolution there should be an icon in the lower right corner of the resized picture. You can also turn off image auto resizing in Tools->Internet Options->Advanced->Multimedia and then clearing the `enable automatic image resizing` checkbox. Or you could get yourself a real browser, ;).
For your reference, a 440 Hz sine wave at full amplitude. The waveform is graphed with time on the x-axis and the intensity of the signal on the y-axis. The energy is represented on a linear scale from 1 to -1 representing the 16 bits per sample of available data space/headroom in a standard CD quality recording.
This spectrograph is mapped
differently than the spectra taken from the MMW track. The 440 Hz
sine wave contains less information, therefore its details are more
easily visible when the FFT window is widened and the y scale is
reduced from 22.5 to 8 KHz.
The signal's spectrum is graphed with time on the x-axis and
frequency on the y-axis. The intensity of the energy is
represented through the use of color ranging from the grey
background (0 amplitude) to blue (low intensity) to red (higher) to
white (max amplitude).
Here's the same sine wave
amplified by 1 dB. About a dozen samples have been
chopped off of each peak and trough. The flat peaks and troughs at unity gain (1 or -1) represent the portion of the signal that can not be contained in 16 bit samples.
This is what happens when the recording levels are
set too high. Once lost the data is gone forever.
Trying to fix a clipped signal requires some involved
audio voodoo to reduce the amplitude, remove the white noise
distortion caused by the flat peaks and reconstruct the lost
information through interpolation.
When in doubt, it is always better to increase headroom than
to risk level settings that are too hot. Any process used to
`fix` a clipped signal will degrade sound quality more than the
reduced signal (and therefore reduced data in the recorded track)
that results from leaving the proper amount of headroom.
The whole waveform of d1t03.
The whole
spectrum.
Remember that this signal is graphed
differently than the reference spectrum above. Here
the y-axis ranges from 0 to 22.5 KHz. Additionally,
the FFT window in this spectrograph is 8 times
narrower.
The main clipped section @ ~5:09 is highlighted in grey.
Closeup of main clipped section -- waveform.
Closeup of main clipped section -- spectrum. Note the red peaks in the high frequencies (say, 16-22.5 KHz) @ 5:09.30 and 5:09.70, especially in the right channel.
Same section, digitally amplified by 5 decibels.
The spectrum of the section amplified by 5 dB. There is an overall increase in intensity, most notably in the high frequencies.
Amplified by 10 dB.
The spectrum amplified by 10 dB.
Sample level zoom of the first clipped trough in the
section. The circled trough has one sample clipped from the bottom.
This recording was made by including some
sort of hardware limiter in the signal chain. When it detects a
certain number of clipped samples in a row it quickly
lowers the recording levels then returns them
to their previous setting when the source's volume is reduced.
While this is convenient, purists find that
brickwall limiters color the sound in a way
that they prefer to avoid through careful setting and monitoring of
their recording levels.
Sample level zoom of first clipped section amplified by 10 dB. There are several peaks and troughs with multiple sample clips in this section. The distortion was audible on this track; even on my little computer speakers.