Reproducing accurate and perceptually realistic spatial audio for augmented and virtual reality (AR/VR) requires the headphones to have a flat frequency response. This can be achieved by equalizing the headphone transducers’ output given the transfer function between the transducer and the human ear, referred to as Ear Acoustic Response (EAR). EAR is unique to every individual and is a function of the transducer characteristics, the user’s anthropometric features (e.g. ear and head shape) and the interactions between the two. This paper proposes a novel method to infer the EAR given the ear features of any listener using a probabilistic framework and a sub-sample of the population as prior. We introduce an approach to assess the level of personalization achieved and benchmark the improvements delivered by the proposed algorithm relative to a generic solution.

We propose a novel approach for blind room impulse response (RIR) estimation systems in the context of a downstream application scenario, far-field automatic speech recognition (ASR). We first draw the connection between improved RIR estimation and improved ASR performance, as a means of evaluating neural RIR estimators. We then propose a generative adversarial network (GAN) based architecture that encodes RIR features from reverberant speech and constructs an RIR from the encoded features, and uses a novel energy decay relief loss to optimize for capturing energy-based properties of the input reverberant speech. We show that our model outperforms the state-of-the-art baselines on acoustic benchmarks (by 17% on the energy decay relief and 22% on an early-reflection energy metric), as well as in an ASR evaluation task (by 6.9% in word error rate).

Although finite-difference time-domain (FDTD) simulations are extensively used in acoustics, studies assessing the reliability and the accuracy of the implemented method are scarce. Moreover, the operational validity of a simulation method is context-dependent. Motivated by the context of head-related transfer function (HRTF) prediction, this work presents a verification procedure for FDTD-simulated HRTFs from a simplified model of a human head and torso which comprises two spheres and is known as the snowman model. The analytic solution required by the code verification is computed with the multipole reexpansion technique and used to compute convergence rate estimates. A solution verification process follows in which asymptotic predictions are computed. Results from the code verification show scattered convergence rates which attained the expected first-order accuracy at frequencies below 1 kHz when a linear regression model was used as the estimation procedure. Results from the solution verification show that the asymptotic predictions are accurate up to 10 kHz, after which bias is observed. Assessing the accuracy of the employed solution verification procedure revealed that the absolute difference between the$\log$-magnitude of the asymptotic predictions and this of the analytic solution is within 1 dB up to 11589 Hz across the HRTF directions considered.