TAU Urban Acoustic Scenes 2019 Mobile: 10-Second Audio Segments from 10 Urban Environments

46 hours of audio data comprise the TUT Urban Acoustic Scenes 2019 Mobile development dataset. It consists of 10-second audio segments recorded in 10 different acoustic scenes, such as airports, metro stations, and parks, using three simultaneous recording devices. The dataset was created by Toni Heittola of Tampere University.

Use Cases

• Train acoustic scene classification models based on 10-second audio segments from 10 defined urban scenes.
• Benchmark multi-device audio recording and synchronization techniques based on parallel recordings from three devices.
• Develop models for urban soundscape analysis based on labeled audio from environments like streets, parks, and public transport.
• Research domain adaptation in audio AI using data captured simultaneously by different microphone devices.

Strengths

• Contains 46 hours of labeled audio data, providing a substantial corpus for model training.
• Includes parallel recordings from three different devices (A, B, C) for 108 segments, enabling device robustness studies.
• Each of the 10 acoustic scenes has 1440 segments (240 minutes) recorded with the primary device, ensuring balanced class representation for core data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• The geographic origin of the recordings is not specified, which may limit understanding of potential regional acoustic biases.

Provenance

Source

Tampere University

Collection Method

Audio recordings made with three devices capturing simultaneously in various urban locations.