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

TL;DR

The paper tackles robust room impulse response estimation in uncontrolled environments by using music as excitation and formulating the problem as time-frequency L1-norm regression. It proposes AnyRIR, solved efficiently with IRLS and LSQR/LSMR, and aided by FFT-based convolution and an excitation-measurement equalization preconditioner. Across 50 synthetic trials and real-world measurements, AnyRIR outperforms traditional L2-based and frequency-domain deconvolution, particularly under non-stationary noise and codec mismatch. This work enables accurate, non-intrusive RIR estimation for AR/VR and related audio applications in real-world, noisy settings.

Abstract

We address the problem of estimating room impulse responses (RIRs) in noisy, uncontrolled environments where non-stationary sounds such as speech or footsteps corrupt conventional deconvolution. We propose AnyRIR, a non-intrusive method that uses music as the excitation signal instead of a dedicated test signal, and formulate RIR estimation as an L1-norm regression in the time-frequency domain. Solved efficiently with Iterative Reweighted Least Squares (IRLS) and Least-Squares Minimal Residual (LSMR) methods, this approach exploits the sparsity of non-stationary noise to suppress its influence. Experiments on simulated and measured data show that AnyRIR outperforms L2-based and frequency-domain deconvolution, under in-the-wild noisy scenarios and codec mismatch, enabling robust RIR estimation for AR/VR and related applications.

Figures (7)

• Figure 1: (Top) A noisy environment, such as cafe, where music is playing and is reverberating, and (bottom) usage of AnyRIR to estimate the RIR with the help of existing music recognition algorithm.
• Figure 2: The proposed AnyRIR method estimates the RIR by minimizing the time-frequency residual between the recorded music and the clean music convolved with the previous RIR estimate. The main tools used are the IRLS and LSMR methods.
• Figure 3: Number of LSMR iterations required for convergence with and without EQ preconditioning.
• Figure 4: (Left) Music signal convolved with an RIR with added stationary and non-stationary noise: $\bm{y}$. (Center) Original non-stationary noise. (Right) Weights $\bm{w}$ computed by AnyRIR to suppress the bins containing the non-stationary noise.
• Figure 5: EDCs from different deconvolution methods compared to the ground truth, showing the superiority of AnyRIR.