TAU Urban Acoustic Scenes 2020 Mobile: 64 Hours of Urban Audio

TUT Urban Acoustic Scenes 2020 Mobile development dataset contains 64 hours of 10-second audio segments recorded in urban environments. The data is organized into 10 distinct acoustic scenes, including airports, shopping malls, metro stations, and various street and transportation settings. Version 2 of the dataset includes corrections for synchronization issues and the removal of incorrect segments.

Use Cases

• [Urban sound classification] based on [10 acoustic scene labels]
• [Model development for audio event detection] based on [10-second audio segments from multiple devices]
• [Benchmarking machine learning algorithms] based on [a standardized, labeled acoustic scene dataset]
• [Research on urban soundscapes] based on [recordings from diverse public spaces and transport modes]

Strengths

• Contains a substantial 64 hours of total audio data for model training and evaluation.
• Provides clear, multi-class labeling across 10 distinct urban acoustic scenes.
• Includes a corrected version (Version 2) addressing synchronization and data quality issues for 118 segments.

Limitations

• Key metadata such as row count, file size, and specific recording dates are not provided in the available sources.
• The dataset's spatial coverage (specific cities or countries) is not explicitly stated.
• The 'Last updated' field is consistently marked as unknown across all platform entries.

Provenance

Source

Toni Heittola, Tampere University

Collection Method

Likely contains audio recordings gathered from mobile devices in various urban locations.

Geography

Urban environments, but specific locations are not detailed.