DeLTA: Urban Soundscape Noise Annoyance Ratings from 1,221 Participants

July 2021 data collection includes 2,980 15-second binaural audio recordings from urban public spaces in London, Venice, Granada, and Groningen. Andrew Mitchell at University College London created this dataset via a remote listening experiment hosted on Gorilla Experiment Builder, where 1,221 participants identified sound sources and rated annoyance. The resulting tabular frame contains 2890 recordings with consensus sound source labels and average annoyance ratings.

Use Cases

• Train audio classification models based on 24 labeled sound sources present in urban recordings.
• Predict human annoyance ratings (1-10 scale) based on acoustic features of binaural urban soundscapes.
• Study the relationship between identified sound sources and perceived annoyance using the consensus label approach.
• Benchmark models for urban sound event detection using the binaural, multi-city recordings.

Strengths

• Includes 2,980 high-fidelity binaural audio recordings from four distinct European cities.
• Annotations derived from a large, controlled human study with 1,221 participants.
• Tabular summary provides a 2890 by 24 data frame with consensus labels and average annoyance ratings.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Audio files used in the experiment were MP3s; higher quality WAV files require a separate request.
• Last update date is unknown; freshness unverified.

Provenance

Source

International Soundscape Database, processed by University College London.

Collection Method

Remote listening experiment via Gorilla Experiment Builder, distributed to pre-registered participants on Prolific.

Time Range

Data collected between July 5th and July 23rd, 2021.

Geography

Urban public spaces in London (UK), Venice (Italy), Granada (Spain), and Groningen (Netherlands).