L'acoustique de la parole en hautes fréquences

At LMSSC, Paris, January 17th 2019, 1 p.m.

Rémi Blandin
Post-doctorant, Institute of Acoustics and Speech Communication (IAS), Technische Universität Dresden (TUD), Germany

For historical and technical reasons, the study of speech has long been limited to a narrow frequency range comprised between 0 and 5 kHz. Nowadays, with the progress of research and technology, the high frequency speech spectrum above 5 kHz is gaining more interest. This implies new challenges for the physical modelling of speech production, since neglected phenomena with a low frequency assumption must be accounted for. Thus, the so widely used plane wave assumption meets its limitation and the whole complexity of the three dimensional acoustic field needs to be modelled. Indeed, at high frequency, anti-resonances and additional resonances that are not predicted by this simple model are observed and predicted by more accurate simulations.

The concept of a unique vocal tract transfer function becomes itself obsolete, since the speech directivity becomes more pronounced and complex. Indeed, in addition to a more complex diffraction of the radiated sound by the head and the torso, the features of the internal 3D acoustic field may influence the directivity. As a consequence, the transfer functions differ considerably depending on the receiving point considered. However, much work is still required to characterise the relationship between the physical phenomena and their consequence on perception.