Better Mixing Through Spectrum Analysis

See what you've been hearing

By Craig Anderton

A spectrum analyzer, whether hardware or software, is a tool that can help analyze a mix, and reveal frequency or dynamics anomalies. It does this by dividing the audible frequency spectrum into hundreds or even thousands of bands (also called "windows") using a process called Fast Fourier Transform (FFT), then displaying the level of each band in a graph or 3D display. This feedback is invaluable in training you to correlate what you hear with your ears to actual data about frequency response and amplitude. Most digital audio editing programs, and even some multitrack hosts (Fig. 1), now include software spectrum analysis tools.

Fig. 1: The AN-879 plug-in for Sonar 6 inserts into a track to provide visual feedback on levels and frequency response.

However, the object is not to aim for a flat response; generally, the highs trail off gently, while what happens in the bass depends on the genre of music. For example, you'll see more bass on a dance mix with a prominent kick drum. A very uneven average bass response may indicate acoustics-related problems—either from room resonances when miking acoustic sources, or from mixing if you're using EQ to compensate for room anomalies of which you're not aware.

Spectrum analyzers are also invaluable for analyzing the spectral response of well-mixed, well-mastered recordings. Compare their curves to yours and see where the differences lie. Differences are not necessarily "bad"; it depends on the music and style. But if, for example, your mixes sound muddy and other CDs don't, investigate what's happening in the bass and lower midrange.

CUSTOMIZING SPECTRUM ANALYSIS RESPONSE

Spectrum analyzers vary greatly in terms of their adjustable parameters, from simple—you can't adjust anything—to multiple parameters that let you customize the analysis and display process. Here are some of the most common parameters.

• FFT size determines the number of samples per band. Higher numbers give better frequency resolution, but require more time to compute the display. When you're looking for frequency anomalies, use a high value, like 16K or 32K. This catches very narrow peaks that might not be seen with smaller FFT sizes.
• FFT overlap sets the amount by which the analysis bands overlap. Higher values (50% and above) provide a more accurate analysis, but increase display computation time.
• Smoothing window determines the analysis algorithm. Different algorithms trade off sharpness of peaks and leakage between neighboring bands (i.e., data in one band influences the ones next to it). Triangular is a compromise between peak sharpness and leakage, Rectangular provides accurate drawing of peaks but high leakage, while Blackman-Harris has little leakage, but the peaks look more rounded.
• 3D vs. 2D shows the information in different ways. 2D shows amplitude vs. frequency, while 3D displays a series of "slices" within the selected region to relate time to frequency and amplitude.
• Range, reference, etc. are parameters that let you adjust the scale, zoom in on specific areas of the graph, change the 0dB reference, etc.
• Linear vs. log response is best set to Log for audio work, as the curve more closely approximates how your hearing works.
• Maximum retains the highest levels reached in each band—like the "peak hold" function found on some VU meters.

Fig. 2: When you call up the parametric EQ in iZotope's Ozone3 mastering plug-in, you'll see a superimposed real-time spectrum, the response curve you've created (the red line), and also a "guide" line (shown in yellow for clarity) that indicates the typical high frequency energy rolloff that occurs in most music.

Different programs do spectrum analysis differently. Some take (or even save) "snapshots," some take an average reading over time, and some show what's happening in real time. A few programs let you compare the input and output spectrum in relation to a signal processing function, or monitor the effects of a particular operation (Fig. 2).

But the bottom line is that even the most basic spectrum analyzers present useful information about your mix. With practice, maybe someday you'll be able to say "This mix needs a slight boost at 12kHz, a major cut at 340Hz, and a minor notch at 50Hz." Until then, use spectrum analysis to learn more about your mixes.