Name: Matan Yifrach Subject: Signal-dependent parametric methods for robust binaural reproduction using BSM method Name: Oded Levi Subject: Binaural Reproduction of Ambisonic Signals Using Sparse Time–Frequency Representations of Reverberant Speech

Name: Matan Yifrach Degree: M.Sc in Electrical Engineering Supervisor: Prof. Boaz Rafaely Subject: Signal-dependent parametric methods for robust binaural reproduction using BSM method. Abstract: Binaural Signal Matching (BSM) is a widely adopted approach for binaural audio reproduction, aiming to recreate the original acoustic scene at the listener’s ears by convolving multichannel microphone signals with a spatial filter. Typically, this filter is designed as the optimal Least Squares (LS) solution to minimize the mean squared error (MSE) between the recorded sound field and the original signal convolved with Head-Related Transfer Functions (HRTFs). However, standard BSM is fundamentally limited by its assumption of a diffuse sound field, resulting in degraded performance in high Direct-to-Reverberant ratio (DRR), and it exhibits little robustness to listener head rotation. In this study, we present an experimental investigation comparing the signal-dependent BSM method to the signal-independent BSM method through the incorporation of spatial signal information via parametric processing. Our approach introduces two methods: directional-BSM (dBSM) and COMPASS. Both methods exploit extracted spatial cues, such as source Direction of Arrival (DOA), DRR, and source variance, to construct an adaptive, signal-dependent spatial filter. These methods are designed specifically for high DRR conditions by incorporating sound field models that explicitly separate the direct and reverberant components of the sound field. We evaluated our framework through extensive experiments, including over 300 Monte Carlo simulations across diverse acoustic scenarios, as well as dedicated listening tests. The proposed parametric BSM methods consistently outperformed BSM (>1 [dB] gain in average), especially in high DRR scenarios (exceeding 4 [dB] gain in average), and demonstrated higher robustness to listener head rotation. Building on these analytic results, the listening tests used as a subjective evaluation reinforced the simulation findings and demonstrated the effectiveness of integrating signal-dependent spatial information into binaural rendering for improved robustness and perceptual realism. Name: Oded Levi Supervisor: Boaz Rafaely Electrical Engineering Master's Degree 3rd Year Subject: Binaural Reproduction of Ambisonic Signals Using Sparse Time–Frequency Representations of Reverberant Speech Abstract: Ambisonics is a widely adopted spatial audio format that enables binaural reproduction with head rotation and individualized HRTFs. In a typical recording of a sound source within a reverberant room, the Ambisonic representation includes a dense mixture of propagating waves corresponding to the direct sound and numerous reflections. This work investigates the perceptual impact of making this representation sparse by modeling room reflections using a limited set distributed over space time and frequency. The study quantifies the degradation or preservation of binaural reproduction quality as the number of reflections is reduced. Results highlight the potential of sparse source representations as efficient proxies for complex room responses, offering a simplified framework for spatial audio coding and for dynamic object-based rendering across space, time, and time–frequency domains.