Source-linked AI summary
Chatty Maps: Constructing sound maps of urban areas from social media data
Luca Maria Aiello, Rossano Schifanella, Daniele Quercia, Francesco Aletta
TL;DR
Urban sound research lacks reliable, high-coverage measurements and perception cannot be captured by noise levels alone. The paper uses geo-referenced social-media picture tags to map urban sounds, emotions, and perceptions across London and Barcelona, finding geographically patterned soundscapes linked to distinct emotional responses and city experiences.
Problem
Urban sound studies lack reliable, high-coverage measurements, while noise level alone inadequately captures people’s varied perceptions.
Method
The paper constructs urban sound categories from tagged data and uses network community-detection algorithms to organize semantically related sound terms.
Results
Social-media tags produced city-scale street-segment sound maps, with sound types geographically aligned to street types and emotions associated with distinct urban sounds.
Takeaways & Limitations
Social-media data can track urban sounds across entire cities without additional infrastructure, supporting soundscape mapping alongside emotional and perceptual analysis.
Takeaways & Limitations
Social-media data have limited spatial and temporal resolution and coverage, producing occasional false positives from misannotated or figurative sound tags.
Abstract
from arXiv · showhide
Urban sound has a huge influence over how we perceive places. Yet, city planning is concerned mainly with noise, simply because annoying sounds come to the attention of city officials in the form of complaints, while general urban sounds do not come to the attention as they cannot be easily captured at city scale. To capture both unpleasant and pleasant sounds, we applied a new methodology that relies on tagging information of geo-referenced pictures to the cities of London and Barcelona. To begin with, we compiled the first urban sound dictionary and compared it to the one produced by collating insights from the literature: ours was experimentally more valid (if correlated with official noise pollution levels) and offered a wider geographic coverage. From picture tags, we then studied the relationship between soundscapes and emotions. We learned that streets with music sounds were associated with strong emotions of joy or sadness, while those with human sounds were associated with joy or surprise. Finally, we studied the relationship between soundscapes and people's perceptions and, in so doing, we were able to map which areas are chaotic, monotonous, calm, and exciting.Those insights promise to inform the creation of restorative experiences in our increasingly urbanized world.
1 Introduction
Urban sound planning has focused mainly on harmful noise, while pleasant sounds, shared terminology, perception, and city-scale coverage remain insufficiently addressed. This study uses geo-referenced social media data to map urban soundscapes and relate them to emotional responses and street perceptions.
- Motivation: Urban planning has largely focused on negative sounds, although pleasant sounds can positively affect city dwellers’ health.Only a few studies have addressed the broader urban soundscape.
- Research gaps: Urban sound research lacks both a shared vocabulary and a scalable way to capture sounds across entire cities.Existing classifications were neither comprehensive nor systematic, while acoustic data collection mainly relied on small-scale surveys.
- Contributions: The study constructs the first urban sound dictionary from tags on 1.8 million geo-referenced Flickr pictures in Barcelona and London.Its six top-level categories are transport, mechanical, human, music, nature, and indoor, and they closely resemble prior manual classifications.
- Contributions: The resulting street-segment sound maps distinguish pedestrian streets, associated with people, music, and indoor sounds, from primary roads, associated with transport and mechanical sounds.The maps were produced for Barcelona and London from picture tags.
- Contributions: The study relates urban sounds to emotions and perceived street qualities using social media tags and soundwalk-based perceptions.Mechanical sounds were associated with fear and anger, while human and music sounds were associated with joy; expected perceptions were inferred from sound tags.
2 Methodology
The method matched sound-related vocabularies to geo-referenced Flickr tags in London and Barcelona, using Freesound for broad coverage and Schafer’s classification for comparability. Freesound terms were further organized into a data-driven urban sound taxonomy based on word co-occurrences.
- Data and vocabulary: The method matched sound-related words to geo-referenced social-media content, using 17M Flickr photos from London and Barcelona and aggregating tags by OpenStreetMap street segment.The photos covered 2005–2015, and street segments were identified from OpenStreetMap.
- Data and vocabulary: Freesound provided 2.2K sufficiently frequent tags from 65K unique tags, retaining 76% of total tag volume.Tags occurring more than 100 times were retained because rarer tags were considered too sparse.
- Vocabulary selection: Freesound matched 2.12M picture tags across 141K London street segments, offering the strongest and broadest coverage among the sound vocabularies.Because 67% of Favoritesounds tags also appeared in Freesound, the analysis used Freesound alongside Schafer’s classification.
- Urban sound taxonomy: Schafer’s words were organized into seven categories, while Freesound words were clustered by location-based co-occurrence using Infomap and further split when clusters were too large.The resulting data-driven taxonomy spontaneously emerged from word co-occurrences and resembled Schafer’s classification.
- Urban sound taxonomy: Freesound was selected for the urban sound wheel because it offered external validity, a richer vocabulary, and fuller representation across sound categories than Schafer’s dictionary.Only the wheel’s top-level categories were used in the subsequent analysis.
4 Emotional and Perceptual Layers
The study extracted emotional and perceptual layers of urban sound from social-media tags and soundwalk data. It linked sound categories to emotions and perceptions, enabling maps of how urban areas are experienced.
- 4.1 Emotional layer: EmoLex classified 6,468 terms into eight primary emotions, which were matched with Flickr tags to compute each street segment’s emotion profile.The emotions were anger, fear, anticipation, trust, surprise, sadness, joy, and disgust.
- 4.1 Emotional layer: Joy was associated with music and human sounds, while traffic was associated with fear, anticipation, and anger.Sadness, together with joy, was also associated with music; all reported correlations were statistically significant at p < 0.01.
- 4.2 Perceptual layer: Soundwalks across 8 Brighton & Hove areas and 11 Sorrento areas produced 342 participant reports measuring five sound categories and eight perceptions on scales.The study involved 37 participants, who listened at selected locations and completed structured questionnaires.
- 4.2 Perceptual layer: Perceptual opposites showed negative correlations, whereas pleasantness and eventfulness were nearly uncorrelated, indicating that these dimensions were orthogonal.The negatively correlated pairs included pleasant versus annoying, eventful versus uneventful, vibrant versus monotonous, and calm versus chaos.
- 4.2 Perceptual layer: Vibrant areas tended to be associated with crowds, pleasant areas with individuals, calm areas with nature, and annoying or chaotic areas with traffic.Conditional probabilities supported the same conclusions, although all were below 0.33.
- 4.2 Perceptual layer: Mapping the strongest sound-perception probabilities showed that trafficked roads were chaotic, while walkable parts of London and Barcelona were exciting.The normalized conditional probabilities were proposed as a basis for aesthetics maps reflecting the emotional qualities of sounds.
5 Discussion
ChattyMaps shares key methodological features with SmellyMaps while relying entirely on Flickr tags, but its soundscape maps are constrained by social-media bias, limited resolution, and partial or ambiguous tagging. The study mitigates these issues through spatial buffering, segment-level sound distributions, and diversity analysis, finding that 28% of Barcelona segments and 35% of London segments had only one tag.
- Relation to SmellyMaps: ChattyMaps and SmellyMaps both used community-detection algorithms to construct taxonomies resembling established field categorizations.Their approaches also shared methods for mapping social-media data onto streets.
- Relation to SmellyMaps: ChattyMaps relied entirely on Flickr tags because picture tags captured geographically salient information more effectively than tweets.This choice followed the finding reported for SmellyMaps.
- Limitations: Social-media data can produce false positives because its limited spatial and temporal resolution misses changing soundscapes, while tags may be misannotations or figurative uses.Soundscapes can change with small variations in location and time, such as turning a corner or comparing day and night.
- Limitations: Partial tagging can omit sound categories present in a street, but this risk decreases as the number of sound tags for a segment increases.Nearby pictures in the example produced different traffic- and nature-related tag profiles.
- Mitigation: The study reduced boundary noise with a 22.5-meter buffer and estimated that around 20-25 tags are needed for high-confidence sound profiles when validated against official air-quality data.It also associated sound distributions rather than individual sounds with street segments and normalized the six-dimensional sound vector to [0,1].
- Sound diversity: 28% of Barcelona segments and 35% of London segments had only one tag and were removed before mapping sound diversity, whose distributions peaked at 1 in both cities, 1.5 in London, and 2.0 in Barcelona.Diversity was computed with the Shannon index using the fraction of tags matching each sound category.
6 Conclusion
Social media data enabled cheap, city-scale tracking of urban sounds that aligned with expected street-type patterns and noise pollution levels. The approach remains limited by dynamic, context-dependent soundscapes and the difficulty of fully capturing them, motivating future multisensory research.
- Contribution: Social media data enabled effective, inexpensive, city-scale tracking of urban sounds that matched expected geographic patterns across street types and noise pollution levels.The method required no additional service or infrastructure.
- Limitations: Soundscape perception varies with demography, city context, and time, so future work could collect more data and compare social-media and GIS-based models.Examples include noise sensitivity, age, city layout, day versus night, and weekdays versus weekends.
- Limitations: Fully capturing soundscapes may be impossible because the study identifies potential sonic events rather than their complete acoustic environment.The paper likens sonic events to raw ingredients, aural architecture to cooking style, and the soundscape to the finished dish.
- Future research: Future research will unite visual, olfactory, and auditory perceptions in comprehensive multisensory studies of cities, supporting more ecologically balanced soundscapes.The stated goal is harmony between human communities and their sonic environments.