Jean-Jacques Embrechts

Efficient database pruning for large-scale cover song recognition

This paper focuses on cover song recognition over a large dataset, potentially containing millions of songs. At this time, the problem of cover song recognition is still challenging and only few methods have been proposed on large scale databases. We present an efficient method for quickly extracting a small subset from a large database in which a correspondence to an audio query should be found. We make use of fast rejectors based on independent audio features. Our method mixes independent rejectors together to build composite ones. We evaluate our system with the Million Song Dataset and we present composite rejectors offering a good trade-off between the percentage of pruning and the percentage of loss.

Numerical evidence of mixing in rooms using the free path temporal distribution

The ergodic propriety of a room has strong effects on its reverberation. If the room is ergodic, the reverberation can be broken up in two steps: a deterministic process followed by a stochastic one. The late reverberation can be then modeled by a reverberation algorithm instead of more computationally consuming methods. In this study, the free path temporal distribution obtained by ray-tracing is used as an indicator of the rooms mixing: the energetic average of the path lengths is computed at each time step. Ergodic rooms are thus characterized by rapidly convergent distributions. The free path value becomes independent of time. On the other hand, path selection mechanism and orbits are observed in non-ergodic rooms. The transition time from the deterministic process to the stochastic one is also studied through the evaluation of the rooms time constant. It is shown that its value depends only on the mean free path and the boundaries scattering value. An empirical expression is obtained which agrees well with simulations carried out in a concert hall. This transition time from a deterministic model to a stochastic one can be used to speed up the acoustical predictions and auralizations in ergodic rooms.

Prediction and measurement of the random-incidence scattering coefficient of periodic reflective rectangular diffuser profiles

Periodic reflective rectangular diffuser profiles have been previously studied by Embrechts through an analytical method calculating their random incidence scattering coefficients. This paper presents the results of further investigations obtained by real-scale and 1:5th scale random incidence scattering coefficient measurements on these profiles. All measurements were performed according to the corresponding ISO 354 and ISO 17497-1 standards. The measurement results are compared with the calculation results derived by the analytical method. Although some periodic profiles could not be treated by the analytical model, fairly good agreements are obtained between the different approaches. The real scale and 1:5th scale measurement results are also presented and compared.

A new validated method for improving the audiovisual spatial congruence in the case of stereoscopic-3D video and wave field synthesis

While 3D cinema is becoming increasingly established, little effort has focused on the general problem of producing a 3D sound scene spatially coherent with the visual content of a stereoscopic-3D (s-3D) movie. The perceptual relevance of such spatial audiovisual coherence is of significant interest. In this paper, we explain why the combination of accurate sound positioning and stereoscopic-3D images can lead to an incongruence between the sound and the image for multiple spectators. Then, we adapt to s-3D viewing a method originally proposed for 2D images in the literature to reduce this error. Finally, a subjective experiment is carried out to prove the efficiency of the method.