In this comprehensive article some deeper explorations and explanations on this topic are given and at the end a brief but handy definition about audio signal coloration is proposed. Some tips on mixing can be obtained here as well and – by the way – some myth about equalizing audio in the digital domain gets busted.
But first let’s have a closer look into a different domain, the domain of digital image processing. In digital image processing, the fundamental color impression of an image is actually changed by performing some proper DSP maths on parts of the color spectrum of the image as shown in the example diagram above. Typically, a digital image is encoded into a 3- or 4-dimensional color space (like RGB or CMYK) and then each dimension can be manipulated individually over the spectrum. This changes then the overall coloration (and other things like brightness or contrast as well).
Quiet similar, in the audio domain there are two dimensions over the frequency spectrum which can be utilized to alter the audio color impression: the magnitude and the phase response and this is typically done by an equalizer or filter. We are not going to talk here about those drastic phasing effects which are typically introduced by time shift based effects such as chorus and delay, just to name the two. While the impact due to alterations in the frequency magnitude response curve is quiet obvious, altering the phase response might be not (and is often mixed up with other EQ side effects like resonance or ringing of a filter).
So, how does a certain phase alteration actually affect the sound perception of the audio? Of course this depends on the real frequency where the phase alteration (aka phasing, phase shift, phase drift, phase warping, phase distortion) occurs but for the sake of simplicity lets first have a look at the rather general effect caused by an overall and continuous phase shift: Lets assume we have an effect which introduces a continuous phase shift over the entire frequency range but changes nothing else (which can be performed by an Allpass filter – see the figure above). Now we have three scenarios, depending on the amount (degree) of the phase shift:
- Slightest phase shift: Human ear does not perceive anything and thus can’t judge any better or worse sonic quality.
- Some amounts of phasing: The phasing, which causes some drift of higher frequencies in time, can now be perceived by our hearing. This slight displacement in time could be perceived as to be a more sound, less edgy and harsh audio quality and even to have more room / depth impression. (Hint: this are qualities some people hear and associate to the rather positive properties of analog audio processing).
- Larger phase drifts: Larger displacements in time increases this effect and can completely destroy the transients. The signal is perceived as being washy and roomy now and lacks of definition which is not desired in most cases (but can be useful e.g. as part of reverberation processors).
In digital reality, phase shift is not introduced that much over the entire spectrum by our commonly used DSP mixing effects such as phasing EQ’s but has a rather local effect and is hearing wise not that easily to detect since the phasing effect is being concealed by the frequency magnitude change of the EQ. This holds true at least as long as gentle and broad magnitude changes are performed. Though, if rather deep and steep changes are made then also larger and maybe unwanted spectrum displacements are introduced by such an EQ.
This may lead to some serious issues when for example on each and every track most signal resonances were removed by such steep filtering effects which is a common misconception in mixing audio. It leads not only to rather flat and boring signals but also introduces significant phasing issues as described above as long as no linear phase EQ is used (which introduces other problems and is not discussed here) and as an overall result the mix gets fluffy and lacks definition. As a side note, this also shows that the urban legend that cutting the frequency is always prefered opposed to boost some other frequency parts holds not true in general.
In some cases it might be considerably better to gently boost the desired frequencies opposed to deeply cut some unwanted ones and the simple “garbage in, garbage out” law applies here too: If that much and rather deep cutting or filtering in general would be necessary on such a signal then it’s probably better to try to fix this by changing the source, the recording situation or the arrangement. A good excercise is to set up some sound sources plus arrangement where (almost) no EQing is necessary during the mix. As an added sugar, such well-selected/recorded and arranged sound sources typically lead to a way better and much more natural loudness performance in the end. But back to topic.
So, is this phasing that bad and has to be avoided in any case? As hinted earlier, performed in slight or the right doses, phasing can introduce a very nice and pleasant audio signal coloration and is part of the sound that we typically associate with high quality audio processing in the analog domain. In DSP land, these kind of effects can be used to the audio engineers advantage as well by simply utilizing phasing in the right amounts and in the right place of the spectrum. This is implemented e.g. in some audio enhancer circuits which are introducing dedicated phase shifts aimed at the spectrum or specific to the loudness performance of the audio signal.
And even some digital compressors are utilizing this, which answers yet another interesting question and that is if we can also introduce audio signal coloration just by using plain dynamic effects and so the short answer is: Yes we can! This is rather obvious and can easily proved by an audio analyser in some cases while in other cases not. Some compressors e.g. introduce gain reduction dependent phasing which can subtly change the color impression and of course a true multiband compressor is able to perform drastic frequency magnitude changes for obvious reasons.
But even if the dynamic processor does not alter the frequency or phase response in a direct fashion at all it can alter the perceived spectrum just by having implemented a frequency dependent sidechain so that parts of the spectrum will be treated differently then others in respect to their loudness performance. This is also true for transient processors in general when transient information is typically associated to (and treated in) specific frequency ranges.
The phrase “audio signal coloration” could simply be seen and understood as “affecting the perceived tonal spectrum” no matter which phenomena or method actually caused it. A phasing EQ, when properly applied, is a good way to pleasantly color the audio in both dimensions, frequency magnitude as well as phase response. A linear phase EQ colors the audio too but just in one single dimension. Other processors such as compressors or enhancers can potentially take advantage of audio signal coloring as well, not to speak of the time shift based effects such as chorus or delay.