CRITICAL BAND THEORY OF SPECTRAL AUDIO PROCESSING

When there are audio signals present in different bins, each signal is heard independently as long as the signals are loud enough and far enough apart in frequency to stay above and away from an adjacent bin’s asymmetrical masking threshold (see Figure 2).

If an audio signal is soft enough in level or close enough in frequency to sneak under the masking threshold of an adjacent band, that signal is masked. It is rendered inaudible. Even so-called Golden Ears can’t hear it (see Figure 3).

The ability of audio signals to mask each other under certain conditions reveals important human psychoacoustic behaviors that we were able to tap into for on-air audio processing.

Slicing the Pizza

When audio is divided up into numerous frequency bins the energy within each bin is reduced according to how many bins the audio has been is divided into – the more bins there are, the less audio there is in each bin. Probably not intuitive in Figure 1 is another subtle clue: when there are 25 bins or more, the ‘sound of processing’ within an individual bin is also inaudible because it can’t break the ‘in-band, multi-stimulus’ masking rule.

Even though it only requires 25 bins to model the human auditory system, we decided to go a bit further with 31 bins, centered on the standard ISO standard 1/3 octave frequencies for our spectral processing.

More Limiter Bands? Better?

Limiters with only a few bands are often seen in operation with more than 6dB of limiting depth. On the contrary, because a spectral processor has such a small amount of audio energy in each band, it typically needs less than 3dB of control per band.

Such shallow limiting combined with the high number of bands meshes so well with how human hearing works that it makes the spectral processor’s operation remarkably invisible to the ear. Gone are the dense, smashed-sounding audio and other annoying characteristics of limiting. The spectral processor simply doesn’t work that way.

A Better Listening Experience. And a Surprise…

When implemented correctly the spectral processor easily manages the energy of electrical signals without our ears noticing that any work has been done. But here’s another secret that not everyone notices first off: the spectral processor actually uncloaks fragile audio details, the same details that are often turned into incomprehensible mush by other limiter techniques.

In a conventional multiband limiter, the broadness of each limiter band (see Figure 4) allows the act of limiting to affect a large portion of the audio spectrum, also reducing nearby frequencies that likely have no need to be reduced. In those limiters, quieter audio details coexisting in the band but below its limit threshold are also pulled down along with the stronger signals, often to below audibility. Many users find themselves driving the multiband limiter harder and harder trying to get the lost detail back.

Conversely, as Figure 5 shows, each band of the spectral processor is quite narrow —note how little audio spectrum is affected by one limiter band. There is something even more important to notice, which is how the high selectivity of the spectral processor allows nearby audio detail to be completely untouched. It is all still there. This is completely different behavior from the way multiband broadcast limiters with only a few bands work. It sounds a lot different, too!

How Does Spectral Processing Sound?

When a band of the spectral processor reduces its gain, two things happen. One is expected, and one is a surprise. The expected thing is that the level of the signal being limited has been restricted to the band’s limit threshold, just as it would be in any limiter. The surprise is that the act of limiting a signal in one narrow band psychoacoustically (not electrically!) raises the perceived loudness of subtle audio details residing near to, but not inside of, the band in limiting. The mechanism for this effect may need an explanation…

Even though the audio signals in the bands adjacent to the one in limiting have not undergone any modification, our brain decodes it very differently and allows us to hear subtle details in the program material not often heard from other broadcast audio processors. Exactly why this effect is heard is open to speculation. But we can conclude that the mechanism for it is quite simple: The frequency in the band undergoing limiting and the audio frequencies in nearby limiter bands not being limited have undergone a change in their relative gains. Our brain doesn’t notice the effect of limiting because, psychoacoustically, it is constrained to such a narrow band. But there’s a perceived increase in the level of the nearby signals in the nonlimited bands even though their electrical amplitudes have not changed. This is entirely opposite behavior from what limiters with only a few bands do when they carve up huge chunks of the audio spectrum just to limit a single isolated signal.

The magic is that there is no magic, only science. The subtle audio details being revealed by the spectral processor have always been there – they were always part of the song, if you will.

We recently introduced Neuron FM/HD spectral audio processor featuring our patented 1/3 octave band resolution technology above.