Skip to content

urban-noise-uganda-61k

Description

The urban-noise-uganda-61k dataset consists of audio samples representing urban noise environments in Uganda, designed for tasks such as noise classification, audio tagging, and sound analysis. It was collected as part of a large-scale noise mapping initiative across Kampala (five divisions) and Entebbe (four wards), in collaboration with city authorities (Kampala Capital City Authority and Entebbe Municipal Council). Each recording is tagged with a noise category, sound pressure level (SPL), GPS coordinates, and a timestamp — making it, per its authors, the first large-scale urban sound dataset collected in an African city, curated for African urban noise categories.

The dataset is described in a peer-reviewed data descriptor published in Scientific Data (2026), and was featured as a Nature Africa Research Highlight in March 2026.

Languages Covered

Not applicable — this is an environmental/noise audio dataset (traffic, sirens, vendors, construction, etc.), not speech.

Size / Records

~62.7k clips total, split across two configurations: - large — full dataset - small — reduced-size configuration

Coverage

City Areas
Kampala Five divisions (Central, Kawempe, Makindye, Nakawa, Rubaga)
Entebbe Four wards (Central, Kiwafu, Katabi, Kigungu)

Data Collection Methodology

  • Pipeline: Field collectors used a custom mobile application to capture short audio recordings at specific georeferenced locations across residential, commercial, and mixed-use urban areas.
  • Noise sources captured: Traffic, sirens, outdoor vendors, religious gatherings, nightlife/entertainment venues, construction sites, and others.
  • Annotations per sample: Noise category, sound pressure level (SPL, dB), GPS coordinates (latitude, longitude, altitude, accuracy), and timestamp.
  • Audio format: Single-channel (mono), 16 kHz.
  • Taxonomy: Noise categories were developed in consultation with city authorities to reflect noise types relevant to urban planning and noise-complaint resolution in the Ugandan context.

Label semantics

The dataset carries two parallel label columns with different origins:

Column Description
class / class_id Ground-truth label assigned at collection time. Kampala samples were hand-labelled by annotators; Entebbe samples were collected without class labels (class_id = 19, unclassified).
predicted_class / predicted_class_id Label assigned post-collection by the SunEcho urban sound classifier via model inference, applied to all samples including Entebbe.

Kampala samples carry both human and model labels (enabling agreement analysis); Entebbe samples carry only model-predicted labels.

Data Fields

Field Type Description
audio Audio (16 kHz) Raw audio clip
class string Human-annotated sound class (collection time)
class_id int64 Integer ID for human-annotated class
predicted_class string SunEcho model-predicted sound class
predicted_class_id int64 Integer ID for model-predicted class
noise_measurement float64 Ambient noise level (SPL) at time of collection
latitude float64 GPS latitude (decimal degrees)
longitude float64 GPS longitude (decimal degrees)
altitude float64 Altitude above sea level (metres)
accuracy float64 GPS horizontal accuracy (metres)
submitter_id int64 Anonymised data collector ID
region string Collection region (Kampala or Entebbe)
timestamp string ISO 8601 collection timestamp

Intended Use Cases

  • Noise classification and audio tagging
  • Spatiotemporal urban noise mapping
  • Urban planning and noise-complaint resolution research
  • Sound analysis / environmental sensing ML applications

Known Limitations

  • Entebbe samples lack human-annotated ground-truth labels (model-predicted labels only).
  • Coverage is limited to Kampala and Entebbe — not representative of other Ugandan cities or regions.
  • Label taxonomy reflects categories relevant to Ugandan urban planning use cases specifically, which may not map cleanly onto other noise taxonomies (e.g. UrbanSound8K, SONYC-UST).

Why This Dataset Is Unique

Unlike existing benchmarks such as UrbanSound8K and SONYC-UST, each sample includes precise geospatial metadata, timestamps, and sound pressure level measurements, enabling detailed spatiotemporal noise mapping. It is also the first large-scale urban sound dataset collected in an African city, curated for African urban noise categories.

How to Load

from datasets import load_dataset

# Load the large configuration
large_dataset = load_dataset("Sunbird/urban-noise-uganda-61k", "large")

# Load the small configuration
small_dataset = load_dataset("Sunbird/urban-noise-uganda-61k", "small")

Licence

Check the "Files and versions" tab on the HuggingFace dataset page for the exact license — not explicitly stated in the dataset card text; please confirm before publishing.

huggingface.co/datasets/Sunbird/urban-noise-uganda-61k

  • Publication: Nsumba, S., Muhanguzi, T., Ouma, E.N. et al. "Noise mapping and ambient sound recordings of the urban environment in Uganda." Scientific Data 13, 345 (2026). https://doi.org/10.1038/s41597-026-06658-w
  • Nature Africa feature: "Fungus fighting fruit flies, dinosaur tracks, and city soundscapes" (Research Highlight, 12 March 2026) — https://www.nature.com/articles/d44148-026-00056-5
  • Figshare (full dataset + raw files): https://doi.org/10.6084/m9.figshare.30168901
  • Real-time noise map (Kampala/Entebbe): https://noise.sunbird.ai
  • Sunbird AI Environmental Sensing: https://sunbird.ai/portfolio/environmental-sensing/

Citation

@article{nsumba2026urban,
  title     = {Noise mapping and ambient sound recordings of the urban environment in Uganda},
  author    = {Nsumba, Solomon and Muhanguzi, Tibabwetiza and Ouma, Evelyn Nafula and
               Sekalala, Imran and Bainomugisha, Engineer and Mwebaze, Ernest and Quinn, John},
  journal   = {Scientific Data},
  volume    = {13},
  pages     = {345},
  year      = {2026},
  publisher = {Springer Nature},
  doi       = {10.1038/s41597-026-06658-w}
}