TUT Sound Events 2018: Simulated Anechoic Recordings with Spatial Coordinates

Tampere University of Technology's dataset contains simulated anechoic recordings of 11 sound event classes, such as coughing and door knocks, with stationary point sources. It includes three sub-datasets for varying levels of temporal overlap, each with cross-validation splits totaling 240 training and 60 testing recordings of about 30 seconds each. The dataset was created by Sharath Adavanne for research on sound event localization and detection of overlapping sources.

Use Cases

• Training sound event localization models based on spatial coordinates (azimuth, elevation, distance) provided in metadata.
• Developing and evaluating detection algorithms for overlapping sound events based on the sub-datasets with 1, 2, or 3 simultaneous events.
• Benchmarking audio classification models using the 11 distinct sound event classes like 'Keyboard' and 'Phone ringing'.
• Creating synthetic training data for spatial audio processing tasks using the simulated anechoic and impulse response recordings.

Strengths

• Provides spatial metadata (azimuth, elevation, distance) for each sound event, enabling precise localization tasks.
• Includes structured cross-validation splits with 240 training and 60 testing recordings per split, facilitating reproducible evaluation.
• Systematically varies temporal overlap across three sub-datasets (maximum 1, 2, or 3 overlapping events), allowing controlled study of source separation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Data is simulated and anechoic, which may not fully represent real-world acoustic environments with reverberation and noise.

Provenance

Source

Tampere University of Technology (TUT), created by Sharath Adavanne.

Collection Method

Simulated recordings using isolated sound events from the DCASE 2016 task 2 dataset, placed randomly on a spatial grid.