AnyRIR: Non-intrusive Room Impulse Response Estimation in the Wild

Kyung Yun Lee1, Nils Meyer-Kahlen1, Karolina Prawda2, Vesa Välimäki1, Sebastian J. Schlecht3,
1 Acoustics Lab, Dpt. of Information and Communications Engineering, Aalto University, Finland
2 AudioLab, School of Physics, Engineering and Technology, University of York, York, UK
3Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Germany
Paper Code

How can we estimate RIRs in the wild, without playing the annoying sine sweeps and when there are lots of non-stationary sounds, such as speech or footsteps?

Real-world Example

We performed real-world measurement in our lab. RIR was measured using the exponential sine sweep.

Centered Image

Case 1: People talking

EDC

EDC image

Weight computed from the IRLS. Non-stationary noises are detected.

Recorded Signal Original Music
Music * Groundtruth RIR Music * AnyRIR Music * L2 RIR Music * FreqDeconv RIR
Groundtruth RIR AnyRIR L2 RIR FreqDeconv RIR

Case 2: People not talking

EDC

EDC image

Weight computed from the IRLS. Non-stationary noises are detected.

Recorded Signal Original Music
Music * Groundtruth RIR Music * AnyRIR Music * L2 RIR Music * FreqDeconv RIR
Groundtruth RIR AnyRIR L2 RIR FreqDeconv RIR

BibTeX

@conference{Lee2025AnyRIR,
          title = {AnyRIR: Non-intrusive Room Impulse Response Estimation in the Wild},
          author = {Lee, Kyung Yun and Meyer-Kahlen, Nils and Prawda, Karolina and Välimäki, Vesa and Schlecht, Sebastian J.},
          booktitle = {Submitted to ICASSP 2026},
          pdf={https://arxiv.org/abs/2510.17788},
          month = {Aug},
          year = {2025},
          }