Tonal Balance Control,  Broad Mode in Ozone 8.png
October 5, 2017 by Gordon Wichern

Why iZotope Created the Tonal Balance Control Plug-in

Go behind the scenes at iZotope and explore why we developed the Tonal Balance Control plug-in in Ozone 8 Advanced and Neutron 2 Advanced.

What is tonal balance, and how do I tell whether my mix has it? This is probably a question most musicians and audio engineers have asked at one point or another, even if they didn’t use those exact words. Tonal balance refers to how frequencies interact with each other, and is often the main culprit in mixes that don’t translate between listening environments (e.g., a mix sounds great in the studio, but not on a car stereo). A common example of a mix that exhibits poor tonal balance is when you can’t quite hear the vocals, so you turn up the volume, but then the bass becomes overwhelming.

Our hope is that understanding tonal balance can help save you from a non-ideal listening environment and speed up your mixing and mastering workflow. To this end, iZotope created the Tonal Balance Control plug-in, which allows you to visualize spectral information in a unique way while also serving as a remote control for any Ozone or Neutron EQs throughout your session.  

To learn how Tonal Balance Control fits into your audio production workflow, check out this article.

To read about tonal balance from a variety of perspectives—musicians, recording engineers, mixing engineers, and mastering engineers—check this blog out.

In this post we'll explore the technical details of the approach we developed at iZotope to quantify tonal balance.

Rethinking the Spectrum Analyzer

The spectrum analyzer is one of the most important metering tools in any audio engineer's toolkit, and it works by displaying the frequency content of an audio signal typically computed using an algorithm called the Fast Fourier Transform (FFT). While the spectrum analyzer is a great tool for identifying resonant and fundamental frequencies, it provides too much information for analyzing tonal balance. I like to use the analogy of GPS navigation software, where the spectrum analyzer is showing you the equivalent of detailed maps at the street/neighborhood level. To analyze tonal balance we want a zoomed-out view that displays things more at the level of a country, state, or province.

In zooming out from the typical spectrum analyzer, we first need to understand the things a spectrum analyzer is measuring that might complicate or confound our ability to measure tonal balance. The first and most important criterion is that we want the tonal balance meter to be level-independent, i.e., we want to measure the overall shape of the frequency spectrum not how loud or quiet a mix is. Additionally, in a typical spectrum analyzer, the view is dominated by the “peaks” in the spectrum, which correspond to the musical notes being played or sung. This means a song transposed to a different key, will look different on a spectrum analyzer, but in terms of tonal balance we’d want the original and transposed song to be similar (assuming everything else is identical). Finally, a typical spectrum analyzer updates several times per second, while for tonal balance we want something that measures overall frequency content, so it should be averaging on the scale of several seconds or even over the entire track.

Using existing tools, e.g., the spectrum analyzers available in the EQ sections of the Ozone or Neutron plug-ins, you can get a good tonal balance measurement by changing the Spectrum Type to Critical, 1/3 Octave, of Full Octave mode, which can smooth out the peaks from the exact notes being played, and the “Average Time” option can then be set to five seconds or greater.

Spectrum analyzer in Ozone 8 EQ with different settings. Linear frequency scale, real-time averaging (left) and full octave frequency scale, 10 second averaging (right).

Spectrum analyzer in Ozone 8 EQ with different settings. Linear frequency scale, real-time averaging (left) and full octave frequency scale, 10 second averaging (right).

While this is a good way to measure spectral content in a zoomed-out way, what’s missing is a way to decide whether a track is tonally balanced or not, i.e., what target you are shooting for.

What Is Well Balanced?

To answer this question, we started with a collection of thousands of commercially available tracks spanning a wide array of musical styles, and also several examples we considered “poorly balanced.” It seemed natural at first to divide this collection based on genre. Genre is useful when navigating a record store, digital music platform, or radio station dial, but also incredibly imprecise and often contentious. For example, what does alternative or indie rock actually mean?

We started with a broad but imperfect set of genres, and our analysis showed genre labels to not be terribly important in the difference in tonal balance between tracks. Most modern (non-classical) music is surprisingly similar when you look at the average spectrum of a track, a finding consistent with the analysis in this academic study. If I were to describe this shape in words, it would be a large bump below 250 Hz or so, a generally flat mid-range between 250-8000 hz, with spikes typically correlating with the overall key of a song (interesting side note the average spectrum of our entire dataset, exhibited 12 mid-range peaks per octave, a likely product of the 12 tone western music scale), and a steep rolloff above 8 kHz.

Average frequency spectra from a popular “Pop” track (blue) and “Heavy Metal” track (white)

Average frequency spectra from a popular “Pop” track (blue) and “Heavy Metal” track (white)

What we found was actually obvious: the biggest variation in tonal balance can be roughly categorized by the intended listening environment. Music that you might listen to in a symphony hall, e.g, classical/orchestral, tends to have its spectral balance dominated by mid-range frequency bands. Music that you might listen to in a club, e.g., hip-hop or certain kinds of electronic music is extremely bass heavy, and most other modern popular music typically enjoyed over headphones or home stereos is somewhere between these two extremes. We then analyzed the variation over all the tracks in each of the “Modern,” “Bass Heavy,” and “Orchestral” groups in order to provide targets quantifying the tonal balance “typical” of best practice norms.

Creating Your Own Guides

Although the three targets provided with Tonal Balance Control provide rough guidelines of the typical best practices in modern recording, many well-balanced tracks might be outside of these bounds some of the time (during a saxophone solo) or even most of the time (a reggae death-metal polka track). It’s also typical in many mixing and mastering workflows to use a reference track (or tracks) to help craft your sound. For this reason, you also have the option in Tonal Balance Control of creating your own target from a single audio file, or creating a composite target from a folder of files, providing the ability to define your own “genre” targets.

As previously mentioned, the Modern, Bass Heavy, and Orchestral target curves included in Tonal Balance Control were created from thousands of tracks. Thus, creating a comparable target from only a single file represented a bit of a technical challenge. Ultimately, the solution we settled on treats every second of audio in a similar way to an entire track in our large-scale analysis. While these custom target curves can be extremely effective, because they are created on such small amounts of audio, things like a mellow intro, drum solo, breakdown, etc. can potentially have a large impact on the calculated targets. Therefore, if you really are trying to closely match a particular section in a highly variable track (e.g., “Bohemian Rhapsody”), you might want to isolate that section in an audio editor before creating the target curve.

Displaying Tonal Balance

In deciding how to display our target curves with a “zoomed out” spectrum analyzer, our goal was to develop a way to quantify whether a given spectrum was tonally balanced with respect to a target. To achieve this, we went back to the “poorly balanced” mixes in our dataset and tried to understand what separated them from “well balanced” mixes. Comparing the entire spectra didn’t turn out to be very useful, but by grouping the spectra into regions and then comparing those regions to the target, all of the poorly balanced mixes were very different from the well balanced mixes in at least one frequency area. Inspired by the way Bob Katz divides the frequency spectrum in the Carnegie frequency chart from his book Mastering Audio: The Art and Science, we divided the spectra into four regions: (1) Bass, below 250 Hz, (2) Lower Midrange, 250Hz -2 kHz, (3) Upper Midrange, 2-8 kHz, and (4) Treble, above 8 kHz.

The “Broad” view in Tonal Balance Control displays a spectrum analyzer, but with all information condensed into the four frequency regions mentioned above, and the target displays the typical variation expected in each frequency range. This view is extremely useful as a “gut check,” to make sure your mix has no glaring tonal balance issues, however, you shouldn’t be alarmed if you are occasionally outside the targets during intros, outros, solos, etc. If your mix does not appear well-balanced in “Broad” view, we also provide a “Fine” view, which displays a more typical spectrum analyzer view (although things have been level normalized and smoothed in time and frequency to still maintain the zoomed out tonal balance qualities). Fine view can be very useful in deciding where exactly to make an EQ change to better match your target frequency spectrum, and using iZotope’s new cross communication technology, you can control any Ozone or Neutron EQ anywhere in your session without leaving Tonal Balance Control.

Tonal Balance Control Fine View in Neutron 2 Advanced

Tonal Balance Control Fine View in Neutron 2 Advanced

Dynamics and Tonal Balance

Most of our discussion on tonal balance has focused on frequency relationships, however, one characteristic that we saw over and over again in mixes that “needed work” was out of control low-end dynamics. In situations where the dynamic range between frequency bands is highly variable, applying a limiter during mastering will essentially have the effect of changing volume for only the most dynamic frequency range (typically the bass, a.k.a, low end).

For this reason Tonal Balance Control also displays a low-end crest factor meter that measures the difference between the highest sample peak and overall level (e.g., RMS or integrated loudness) in your audio. Mixes with an overly dynamic low end will consistently be past the far left line shown on the meter, and can often benefit from some gentle low-end compression, so final limiting can be much more transparent. This can be achieved with dynamic nodes in a Neutron EQ, which can be controlled inside of Tonal Balance Control, or with your favorite compressor.

Additionally, if your mix has a very small low-end crest factor (at the far right of the meter) you may want to dial back your compression/limiting depending on where you stand in the Loudness Wars.

Wrapping Up

The unique spectrum analyzer, targets, and communication technology present in Tonal Balance Control can be extremely valuable in mixing and mastering workflows, and provide added confidence that your track will “measure up.” However, attempts at achieving a target should never compromise your creative vision. The goal in creating Tonal Balance Control was not to steer anyone towards a specific sound, but to help you achieve your target sound (assuming it even exists) faster. Whether that’s being competitive with today’s top tracks or capturing certain qualities from your favorite reference, hopefully having Tonal Balance Control on your master bus can help get you there.

In the future, advances in psychoacoustics might allow for targets that better account for the human auditory system. Another area that could be exciting is incorporating advances in music information retrieval, where high-level music information such as downbeats, chords, transcriptions, etc. are found automatically using machine learning algorithms. This has the potential to create a more precise meter and provide different metering feedback during things like downbeats or chord changes.