How Flickr Tags Can Reveal Entire Acoustic Atlases of Cities

We are mostly a visual species, which is fair enough. Seeing suits our needs—providing a "uniquely rich," highly efficient means of perceiving our environment. Of the senses, vision arguably offers us with the most information bang for our buck. Naked mole rats we are not.

But sound too has a very large impact on how we subjectively perceive spaces. It can also impact our moods, cognitive functioning, and even our physical health. The soundscapes we live within can wind up mattering a great deal, for better and for worse.

To this end, researchers from Yahoo Labs, the University of Sheffield, Bell Labs, and University of Turin have crafted a new system for characterizing and mapping urban soundscapes using data harvested cheaply from existing social media. The result, dubbed Chatty Maps, is described in a paper posted to the arXiv preprint server.

As the researchers note, most efforts toward urban sound mapping have focused on "bad" or unpleasant sounds, aka noise. This makes sense. Noise has been shown to have all sorts of negative health consequences, and, well, people complain about noise. A result of this emphasis, according to the researchers, is that current sound monitoring efforts in cities are targeted at specific sound sources, such as freeways, rail lines, and industrial activity. Sound mapping as currently practiced isn't very holistic.

"Pleasant sounds have been left out from the urban planning literature, yet they have been shown to positively impact city dwellers' health," the paper explains. "Only a few researchers have been interested in the whole 'urban soundscape.'"

"In the World Soundscape Project, for example, composer Raymond Murray Schafer and colleagues defined soundscape for the first time as 'an environment of sound (or sonic environment) with emphasis on the way it is perceived and understood by the individual, or by a society,'" the authors continue. "That early work eventually led to a new International Standard, ISO 12913, where soundscape is defined as '[the] acoustic environment as perceived or experienced and/or understood by a person or people, in context.'"

Since that definition, however, urban soundscaping hasn't exactly taken off as a concept. Some large part of this, according to the study, is that there's no real shared vocabulary that can be drawn on for urban sounds. There's some basis for classifying urban sounds in a machine learning sense, but it doesn't employ a formal taxonomy.

Then there's the general problem of linking sounds with perceptions. "So far the assumption has been that a good proxy for perceptions is noise level," the researchers note. 'But perceptions depend on a variety of factors; for example, on what one is doing (e.g., whether (s)he is at a concert). Therefore, policies focusing only on the reduction of noise levels might well fall short."

Finally, there's the problem of capturing urban sound environments "at scale" or to any degree that approaches comprehensive. Thus, we're mostly left with acoustic snapshots.

So, the task of Chatty Maps is a big one: mapping the soundscape of entire cities using geo-referenced social media data. The first task was to develop a sound dictionary—a collection of sound-related terms harvested from tags on 1.8 million geo-referenced Flickr images from London and Barcelona. The next step was just taking those tags and actually plotting them out on a map. The results were pretty obvious, with pedestrian-heavy areas revealing soundscapes characterized by music and people, while roads were dominated by mechanical and transport-related sounds.

The corresponding emotions were also fairly obvious: "For the first time, we studied the relationship between urban sounds and emotions," the paper notes. "By matching our picture tags with the terms of a widely-used word-emotion lexicon, we determined people's emotional responses across the city, and how those responses related to urban sound: fear and anger were found on streets with mechanical sounds, while joy was found on streets with human and music sounds."

The final task was relating soundscapes with people's perceptions of the corresponding environments. The researchers accomplished this by conducting "soundwalks" in which urban residents were asked to identify sound sources and report them along with their subjective impressions. And so, using this soundwalk data, they were able to add a further perceptual dimension to their sound maps.

There are quite a few obvious limitations here. A big one is that we're drawing generalizations from some very subjective data. I for one love the sound of a passing train, but imagine that particular response would get washed out in Chatty Maps. And there's also the whole issue of soundscapes changing through time: day to night, weekday to weekend, season to season.

"No matter what data one has, fully capturing soundscapes might well be impossible," the paper concludes. "Our work has focused on identifying potential sonic events. To use a food metaphor, those events are the raw ingredients, then the aural architecture (which comes with the acoustic properties of trees, buildings, streets) is the cooking style, and the soundscape is the dish."