SONYC-UST: Urban Sound Tagging Data from a New York City Sensor Network

23 fine-grained sound classes, grouped into 8 coarse classes, were annotated by citizen scientists for urban noise monitoring. The dataset contains 10-second audio recordings from over 50 sensors in the SONYC network, which has collectively gathered the equivalent of 37 years of audio data. A subset of this data, split into training and validation sets, was released in March 2019 by researchers from New York University and other institutions.

Use Cases

• Training multilabel audio classifiers based on the 23 fine-grained sound classes.
• Developing urban noise monitoring systems based on real-world acoustic sensor network data.
• Benchmarking models for the DCASE Urban Sound Tagging task based on the provided validation set.
• Studying citizen science annotation reliability based on the three-volunteer-per-recording annotation process.

Strengths

• Annotations were validated for reliability using a two-stage procedure with two independent annotators for the validation set.
• All recordings are standardized 10-second clips recorded with identical microphones and gain settings.
• Data originates from a real-world, long-term sensor network with the equivalent of 37 years of audio collected.

Limitations

• Row count and total number of provided audio clips are unknown, which may limit suitability assessment.
• Time and location metadata are omitted for privacy, limiting spatial and temporal analysis.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Music and Audio Research Lab, New York University, and collaborating institutions.

Collection Method

Audio recorded from the SONYC urban acoustic sensor network; annotations gathered via the Zooniverse citizen science platform.

Time Range

Temporal coverage is not specified in the provided metadata.

Freshness

Last update date is unknown; freshness unverified.

Geography

Recordings are from New York City, but specific locations are not included for privacy.