VAST: The Virtual Acoustic Space Traveler Dataset

Abstract

Abstract. This paper introduces a new paradigm for sound source localization referred to as virtual acoustic space traveling (VAST) and presents a first dataset designed for this purpose. Existing sound source localization methods are either based on an approximate physical model (physics-driven) or on a specific-purpose calibration set (data-driven). With VAST, the idea is to learn a mapping from audio features to desired audio properties using a massive dataset of simulated room impulse responses. This virtual dataset is designed to be maximally representative of the potential audio scenes that the considered system may be evolving in, while remaining reasonably compact. We show that virtually-learned mappings on this dataset generalize to real data, overcoming some intrinsic limitations of traditional binaural sound localization methods based on time differences of arrival.

Hearing in a shoe-box: Binaural source position and wall absorption estimation using virtually supervised learning

“…Moreover, we observed that incorporating random diffusion effects in simulations significantly increased the spatial richness of binaural features, improving estimation of all parameters. In a simultaneous companion paper [19], we validated the virtually-supervised learning framework by successfully localizing sources from real-room signals and showed the superiority of the approach over a traditional time-delay-based method. Extensions to speech [6], multiple sources [5] and additional partially-latent variables in GLLiM [20] will also be considered.…”

“…Overall, each pair of generated RIR depends on four parameters: the source's azimuth, elevation and range, and the mean absorption coefficient of walls. The complete dataset is available online at http://theVASTproject.inria.fr (see also [19]).…”

Supervised Contrastive Embeddings for Binaural Source Localization

“…To evaluate the embeddings in varying acoustic conditions, we used the VAST dataset [24] of simulated binaural room impulse responses of a KEMAR dummy head [25,26]. The training set consists of 15 rooms with reverberation time 0.1-0.4 s. Each room contains spherical grids of positions with radii 1, 1.5, and 2 meters, centered at 9 positions.…”

Audio-motor integration for robot audition

“…This mapping is then reversed through Bayesian inversion, yielding an efficient estimator ofz givenτ(ξ). GLLiM was notably successfully applied to supervised binaural sound source localization using either real [16,12] or simulated [22] training sets. Here, a fixed value Q = 50, diagonal and equal noise covariance matrices and equal mixture weights are used in all experiments (see [13] for details on the GLLiM method).…”