A great deal of research underpins the effectiveness of sonic branding when it’s done well, and a great many examples of successful sonic branding exist. But for marketers, the process behind the creation of sonic branding has been something of a black box. Specialised agencies staffed by composers and musicians listen to a brief, sequester themselves in studios, apply arcane processes, and emerge later with music that they insist satisfactorily represents a brand’s values or aspirations.
Audio branding is also seen, especially in Asia, as less than essential. It's something that big global brands can spend untold millions on—Mastercard partners with Linkin Park, Netflix hires no less than Hans Zimmer—but many brands see it as a luxury and suspect that it’s too expensive to consider.
A couple of recent projects have drawn back the curtains on the creation of audio branding in ways that could make marketers open their ears to this often underappreciated aspect of branding.
Amsterdam-based music agency MassiveMusic, along with a company called SoundOut, which got its start helping record labels analyse the appeal of popular songs, recently unveiled a tool called MassiveBASS. The "BASS” stands not for chunky low-end sounds, but for “Brand Attribute Sound System".
To create the tool, MassiveMusic had 500,000 people around the world listen to more than 300 musical clips from its archives and choose the attributes—trusted, authentic, cool, desirable, bold, disruptive, fun—that they thought fit best with each particular piece. Each piece of music was rated by around 1500 people.
The company also tagged each piece of music with descriptive information, such as genre, instrumentation, tempo, musical key signature, and so on. The science behind all this gets a little hairy (see the box below if you want details), but in the end, MassiveMusic says it has mapped “the emotional DNA of music”, allowing it and its clients to better home in on sounds and styles that are proven to align with the traits a given brand wishes to portray.
The tool—and especially the research behind it—appears to represent an unprecedented effort in this area. Many music agencies (Adelphoi, Sonic Minds, The Sound Agency, Audiobrain, and Made Music Studio to name a few) talk about their ability to create appropriate music for a given brief. Another, Audiodraft, claims to be able to match brand attributes to music based on data from "20,000+ music producers in 120+ countries”. But given its scope, MassiveMusic’s effort appears to us to be, well, the most massive validation effort in use by a branding agency.
A publicly available beta version of the MassiveBASS tool allows brands to get a small taste of what the real tool does.
Marijn Roozemond, senior creative strategist with MassiveMusic, observes that many brands come into the sonic-branding process saying only that they want something “epic” and unique. MassiveBASS helps to make those initial conversations more fruitful.
“This is a tool for us to kickstart our creative process,” he told Campaign Asia-Pacific while providing a demo. “By no means will we ever say, ‘This is the sound of your brand’. But it gives us guidance in finding the sound of a brand. It's a very playful back end for us to start workshopping with clients. It's very helpful for us to also immediately talk about music and bring music to the table from the get-go, from the first meeting.”
Whereas normal practice in the industry has been to use testing to confirm at the end of the process whether a given composition evokes the qualities the brand desires, MassiveBASS serves to provide a “well-founded"—and audible—starting point.
“With this tool we give ourselves a creative compass that is validated with the opinions of consumers,” Roozemond said. “It's more of setting our dot on the horizon up front, at the creative start, which is validated, so we actually have a better scope of what we want to hit.”
Clients don’t simply select the musical pieces the tool identifies as good matches, of course; the pieces it comes up with are just samples that help brand and agency arrive at a better understanding of how the bespoke work to come should take shape.
“I'm not trying to replace myself with a machine by any means,” Roozemond said. “I'm trying to make better, more informed decisions.”
Adding the tool into MassiveMusic’s process also prods brand marketers into thinking more about what their brand really means. “What I find interesting is that this tool also is quite confronting for clients,” Roozemond said. “If you just give generic brand values, you get generic music. So you’ve got to stand for something if you want to stand out.”
Tamon Fujimi, MassiveMusic’s Tokyo-based director of creative development, said that while the company has found success in Asia, many brands still aren’t really aware of what sonic branding is and how it could help them. The beta tool, which presents a simplified version of the full MassiveBASS tool and lets users explore potential sounds via selecting archetypes, could help spread some understanding, he said.
At the same time, the number of Asia-based brands with broader mindsets is on the rise. “There are tons of brands that are becoming more global, which have the capability to understand a little bit more how branding is not just about the logo or visuals,” Fujimi said. “There's more to it that they can actually do, and that's where we can come in to help them out.”
MassiveMusic purposefully aimed for a global-level understanding of the correlation between music and brand attributes, Roozemond said. So while people from China participated in the survey, the company is making no effort to separate out the sounds and tones that match with brand values in the minds of Chinese consumers, or consumers in any other market.
"We're really focusing on creating identities that work on a global scale,” he said. “It's a pitfall to think that for certain cultures we would need to add certain instruments. I don't think that's the world we live in anymore. I think there's a global sound that resonates with everyone."
What do flowers sound like?
Another European sonic-branding agency that talks up the science behind its process, DLMDD, recently worked with Singapore Airlines on creating a new sonic identity. The result is a suite of symphonic pieces to be played in different settings during the consumer journey (literally), such as boarding, landing and lounge.
But the origin story of those pieces of music is quite interesting. The project in essence translated one of the brand’s visual elements, a batik pattern including flower species native to Singapore, into music.
“It was a really bold, brave and exciting brief from SIA,” Max De Lucia, DLMDD founder and client director, told Campaign Asia-Pacific. "How can we blend art and science to truly translate the visual floral world of the Batik Motif into sound?”
An audio architect, Dominic Murcott, designed and built a unique digital instrument that assigned frequencies (notes) based on the colour of each flower, and used this to generate 14 melodic fragments. A composer, Rohan De Livera, then used these fragments to compose symphonic melodies, which were then recorded by the Budapest Symphony Orchestra.
“None of us, ourselves included as a music agency, had ever done anything like this before," De Lucia said. “What we were trying to achieve was a real world-first for a brand, and we had to be very candid in saying upfront that we had no idea if building an instrument would work. There was a great sense of trust, collaboration and exploration from all parties that allowed for the highest degree of artistry from the composers and creators. That exploratory ethos and ambition, combined with the rigour of DLMDD's sonic branding process, was the magic sauce for the creative result you hear today.”
DLMDD uses proprietary “brand sound benchmarking paradigms” throughout its process to “ensure clients have the sonic science and validity to know they are on the very best direction of travel”, De Lucia said. In this case, not just statistical evidence but also the science of sound itself was "baked into the creative process” from the start.
De Lucia expressed hope that the SIA project will lead to an increase in the use of sonic branding by brands in APAC. “Since the launch of the sound of Singapore Airlines just a couple of weeks ago, we have been blown away by the level of interest in DLMDD and sonic branding from brands across APAC,” he said. “If that's anything to go by, there's going to be many more brands across the region coming to life in sound very soon. On that note, if there's anyone out there who wants to launch our APAC headquarters, the time is now.”
The science underlying MassiveBASS
MassiveMusic provided an in-depth description of the statistical analysis that underpins MassiveBASS, which we have lightly edited and presented below. In essence, the company used statistical techniques to map out the relationships between pieces of music and brand attributes. With that grounding established, it then used some machine-learning techniques to help it analyse any new piece of music, not just the ones tested. Fair warning: You may need a high level of knowledge to follow what follows.
MassiveMusic’s survey of 500,000 consumers collected more than 6 million data points. From that, the company constructed ratings for each brand attribute and each audio asset. For each audio asset, it therefore has all the attribute scores, and conversely for each attribute, it has all the asset scores.
The correlation between two given attributes is computed as the pairwise Pearson's correlation of the asset’s scores for the first attribute, against the asset’s scores for the second attribute. This represents the strength of the linear relationship between the two attributes. In other words, a high correlation between "Happy" and "Joyful" means that the assets that scored high for "Happy" tended to also score high for "Joyful" (and assets that scored low in one tended to score low in the other).
A negative correlation (close to -1) indicates an opposite relationship; assets that score high in one attribute tend to score low in the other. A low correlation (close to 0) indicates a lack of linear relationship; there is no link between the scores of those two attributes.
Next, the company created 2D representations of the correlations, using dimensionality reduction techniques. This involves plotting each attribute as a point in a 250-dimensional Euclidean vector space, whose coordinates are the 250 asset scores for that attribute. Once you see attributes as points in a vector space, you can do math on them. In particular, this allows the company to compute the distance between two attributes.
The proximity of the points on the 2D network plot is determined using multidimensional clustering on the correlation matrix, so attributes that are similar end up close to one another on the graph. The color and strength of the link between them represents their correlation.
Advanced technical analysis
In this graph, numbers in the upper right portion show correlations. For example, ‘Artistic’ and ‘Accurate’ are 60% correlated. The lower left portion shows the scatterplots of the attribute scores. In these plots, each point represents an audio asset (so there should be about 250 points on each plot). The coordinates of the points are based on benchmarked attribute scores. In each plot, points at the top right are audio assets which scored high in both. Points at the bottom left scored low in both. Points at the top left scored high in the y-attribute and low in the x-attribute. As you can see, most of the scatterplots show a clear upward diagonal trend, which indicates a positive correlation. A downward trend would indicate a negative correlation, and a lack of trend would indicate a lack of correlation (a correlation close to 0).
How MassiveMusic can infer 200+ attributes for any new piece of music
Having understood how the attributes relate to each other, the company was able to condense that information and build advanced statistical models to estimate attribute scores for any new asset.
The first step here was to condense the attribute space from 200+ dimensions to only 14, using machine learning and dimensionality reduction. All new assets get tested for these 14 core attributes, which the company says represent the music’s DNA.
Next, using machine learning on its over 6 million data points, the company built statistical models that leverage the relationships between attributes (as shown on the correlation maps) to derive accurate scores for any of the 200+ attributes, based on the 14 core attribute scores.