TAU-NIGENS Spatial Sound Events 2021: Synthetic Scenes with Directional Interference

Spatial sound-scene recordings for sound event localization and detection, developed for the DCASE 2021 Challenge. The dataset contains sound events spatialized using real room impulse responses from multiple rooms, delivered in both microphone array and first-order Ambisonics formats. It is the third iteration from Tampere University, introducing more challenging conditions like a maximum polyphony of 3 target events and directional interference events.

Use Cases

• Training sound event localization models based on direction-of-arrival labels for static and moving sources.
• Developing audio source separation algorithms based on scenes with up to three simultaneous target events.
• Benchmarking detection robustness against directional interference events sourced from 'engine', 'fire', and 'general' sound classes.
• Evaluating model performance across different spatial recording formats (microphone array and first-order Ambisonics).

Strengths

• Uses real spatial room impulse responses captured in multiple rooms with various shapes, sizes, and acoustical properties.
• Includes both static and moving sound sources, with trajectories provided for moving sources.
• Introduces directional interference events, a main new challenging property for the 2021 iteration.
• Provides data in two spatial recording formats (MIC and FOA) for flexibility in model input.

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Tampere University, created by Archontis Politis for the DCASE 2021 Challenge.

Collection Method

Synthesized from the NIGENS general sound events database, spatialized using real captured room impulse responses.