Is there a primer for correctly understanding spectrograms? I use Spek
left-right is time
bottom to top is increasing frequency.
The kicker is that in order to analyze the sound for frequencies, you <em>have to</em> chop the sound wave up into little bits of equal time, since frequency is necessarily encoded in the sound on the same axis as time. This means you have a tradeoff between time accuracy and frequency accuracy. If you have foobar2000, fire up the Spectrogram visualizer, (not to be confused with Spectrum Visualization), right-click on it and have a look at FFT Size. If you increase the size, you get better resolution for lower frequencies, which is somewhat beneficial if you set the Scale to Logarithmic.
You can never get to perfection, but the nature of the eye is such that you can get away with a lot on a monitor and you don't actually need perfection to get a satisfying picture.
The resolution of a spectogram is usually a good deal lower than what your ears can perceive, so a spectrogram cannot be used as conclusive evidence for whether you can or cannot hear something. More specifically, you can't show people a spectrogram and claim that the music sounds crappy or good. You have to listen.
I've always found spectrograms highly interesting because they visualize sound in the most human-intuitive way.
Is there a primer for correctly understanding spectrograms? I use Spek
I'm not a fan of spectrograms at all.
I agree that they generally seem to lack the resolution required to obtain useful information about sounds.
IME, the issue is when people fail to take into account how human hearing works when applying these graphs to the human experience called sound.
If by spectrogram it is meant representations like the Spectral View is Cool Edit/Audition, I take issue with the "not very useful" viewpoint. No doubt it depends on what one wishes to do. I've been working with cleaning up recordings, mostly from LPs and cassettes, for quite some time. I work in Spectral View almost all the time. From experience, I can tell a great deal about the aspects of the sound that interest me, far more than from waveform view.
I do agree that such a representation tells one nothing about how the audio sounds in any absolute sense; you can't tell good from bad music except in some extreme cases, but there is much information available there.
Similarly, spectrograms are very useful to developers. E.g. an implementation bug that causes a phase discontinuity can be picked out in an instant from a spectrogram of a test signal perhaps several minutes in length.