Using a Binaural Spatialization Tool in Current Sound Post-Production and Broadcast Workflow

This paper describes an experiment designed to study the di erences between 5.1 audio played through loudspeakers and headphones in binaural, and between compressed, and uncompressed audio les. Differences in terms of spatial impression and of overall quality of the sound have been studied. This experiment was made in the context of "NouvOson", a Radio France website launched in March 2013 (http://nouvoson.radiofrance.fr/ ), where audio contents are available online both in native 5.1 and processed in binaural using SpherAudio software by Digital Media Solutions. It also concerned the BILI Project, dealing with BInaural LIstening, involving Radio France, France Televisions and DMS. Binaural processing theoretically allows the reproduction of 3D sound when listening through headphones; however, this technology still faces issues. These are not only due to the actual limits of research and development, but also to the way we listen to and localize sounds. This experiment has shown that spatial charasteristics, as well as timbre of the sound are modi ed. Besides, no real di erence in the listener's perception has been found between binaural uncompressed les and AAC 192 kbps as well as MP3 192 kbps les. INTRODUCTION Binaural technologies are based on the fact that we localise sounds thanks to our audio filters called HRTFs (Head-Related Transfer Functions, [1], [2]), which characterise the way sound is modified by the morphology of the head and the outer ear, depending on the sound source position. This implies that HRTFs are different for each one of us. The basic principle of a binaural synthesis is to convolve a sound by the pair of HRTFs corresponding to any desired position (see Begault [3], pp. 95-100 and Nicol [4], pp. 119-125). When the resulting stereophonic signal is listened to with headphones, the sound should be perceived at this position. As this is only a perceptual effect involving virtual sound sources, there are some limitations, e.g. non-individualization of the HRTFs (see [4], pp. 132-139), spatial interpolation of the HRTFs where no data is available (see [5] and [6]), Fig. 1: The workflow used by Radio France for NouvOson. and inability to move the head relative to the sound. Indeed, head movements allow us to determine the position of a sound more precisely by ensuring a dynamic combination of filters (see Begault [3], pp. 39-40). For example, Wightman and Kistler have studied the resolution of front-back ambiguity with head movements [7]. AES 57TH INTERNATIONAL CONFERENCE, Hollywood, CA, USA, 2015 March 6–8