UPV_RIR_DB: A Structured Room Impulse Response Database with Hierarchical Metadata and Acoustic Indicators
Researchers release a structured database of 18,976 real-world room impulse responses with hierarchical metadata.
A consortium of Spanish university researchers has published UPV_RIR_DB, a meticulously structured database of real-world Room Impulse Responses (RIRs) designed to advance AI-powered audio processing. The dataset, hosted on Zenodo, contains 166 multichannel RIR files measured across three rooms at the Universitat Politècnica de València, representing a total of 18,976 individual impulse responses. Each file captures sound propagation for multiple source-receiver pairs across 25 cm² zones, mimicking personal sound areas. Crucially, the database adopts a hierarchical organization where directory structure and metadata files jointly describe the exact experimental context, hardware used, and precise spatial coordinates.
Beyond raw audio data, UPV_RIR_DB includes comprehensive metadata with acquisition parameters and key acoustic indicators like reverberation time, enabling traceable and reproducible analysis. This structure provides a consistent framework for storing, inspecting, and reusing real RIR measurements, ensuring compatibility with both MATLAB and JSON-based workflows. For AI developers, this means a reliable benchmark dataset for training and validating models for speech enhancement, audio source separation, virtual acoustics, and augmented reality audio, moving beyond synthetic simulations to real-world acoustic complexity.
- Contains 18,976 single impulse responses from 166 multichannel files measured in 3 real rooms.
- Features hierarchical metadata with spatial coordinates, hardware specs, and acoustic indicators (e.g., reverberation time).
- Publicly available on Zenodo, providing a traceable, real-world benchmark for AI audio model development.
Why It Matters
Provides a high-quality, real-world dataset to train more robust AI models for speech enhancement, virtual acoustics, and AR/VR audio.