TAU Moving Sound Events 2019: Ambisonic Recordings with Moving Sources

A 2019 dataset from Tampere University contains real-life first-order Ambisonic recordings of moving sound events. It consists of three sub-datasets with up to three temporally overlapping events, each with 240 training and 60 testing recordings of about 30 seconds. The dataset was synthesized using impulse responses from a university corridor and sound events from the UrbanSound8K dataset.

Use Cases

• Training models for moving sound event detection based on Ambisonic audio recordings.
• Developing spatial audio source localization algorithms based on azimuth and elevation metadata.
• Researching overlapping sound event separation in reverberant environments based on the multi-overlap sub-datasets.
• Benchmarking acoustic scene synthesis methods based on real-life impulse responses and constant-velocity source trajectories.

Strengths

• Recordings are in first-order Ambisonic format with spatial metadata including azimuth, elevation, and angular velocity.
• The dataset provides three distinct sub-datasets with up to one, two, or three temporally overlapping sound events.
• Each sub-dataset includes structured cross-validation splits with 240 training and 60 testing recordings.
• Sound events are drawn from 8 foreground classes of the established UrbanSound8K dataset.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Tampere University, created by Sharath Adavanne.

Collection Method

Recordings generated by collecting impulse responses in a university corridor using an Eigenmike array and a moving Genelec loudspeaker, then synthesizing scenes with sound events from UrbanSound8K.

Time Range

2019

Geography

Recorded in a corridor at Tampere University.