TUT Urban Acoustic Scenes 2018: 10-Second Urban Audio Segments

8640 ten-second audio segments capture 10 distinct urban acoustic scenes, totaling 24 hours of audio. Each scene, such as airport, shopping mall, and various transportation modes, is represented by 864 segments (144 minutes). The dataset is designed for developing and benchmarking machine listening models in urban environments.

Use Cases

• [Acoustic scene classification] based on [10-second audio segments from 10 urban scenes]
• [Urban sound event detection] based on [recordings from environments like metro stations and pedestrian streets]
• [Benchmarking audio embedding models] based on [labeled, fixed-length audio clips]
• [Training models for environmental sound recognition] based on [scenes including traffic, parks, and public transport]
• [Studying urban soundscape characteristics] based on [recordings from diverse public spaces]

Strengths

• Contains a balanced 864 segments per class, enabling robust model training and evaluation.
• Provides a substantial total of 24 hours of annotated audio data for urban acoustic research.
• Focuses on a specific and well-defined set of 10 common urban acoustic scenes.

Limitations

• Column names and specific audio file formats are not detailed in the provided metadata.
• The dataset's 'development' designation suggests it may be part of a larger collection, with potential evaluation sets missing.
• Metadata on the last update is unavailable, limiting assessment of freshness.

Provenance

Source

Tampere University, authored by Toni Heittola.

Collection Method

Likely contains field recordings gathered in urban environments.

Time Range

2018 (inferred from title).

Geography

Urban areas, specific locations not stated.