13th Speech in Noise Workshop, 20-21 January 2022, Virtual Conference 13th Speech in Noise Workshop, 20-21 January 2022, Virtual Conference

P10 Does speech intelligibility depend on the spatial appraisal of the sound source in diffuse noise?

Nicola Prodi, Chiara Visentin, Matteo Pellegatti
Department of Engineering, University of Ferrara, Italy

(a) Presenting
(b) Attending

In order to build up a spatial impression of the speech source the listener needs to extract several monaural and binaural cues from the signal. When the room impulse response is analyzed to this scope, it is known that some relevant cues are obtained by the specific elaborations of the early reflections, for instance exploiting the correlational mechanisms supporting binaural hearing. On the other hand, it is also known that the energetic balance between the early reflections and the reverberant tail has a strong impact on speech intelligibility, and that the ratio of the direct sound over reflections together with the direction of arrival of the reflections greatly affect spatial perception too. Thus, even in reverberant quiet conditions the interplay of spatial percepts and speech intelligibility is a complex issue, which is not much addressed in the literature. When diffuse noise is added on speech, it is unclear whether and how the spatial appraisal of the source (e.g. percepts of distance, width, focus and envelopment) is altered, nor it is clear if this would affect speech intelligibility. In this work impulse responses with specular or scattered early reflections are used to manipulate the spatial percepts while keeping the energetic balance between early sound and the reverberant tail fixed according to the concept of sound clarity. Two different reverberant tails are employed, and several virtual sound fields are created. To this aim anechoic speech is convolved with suitable impulse responses; a quiet and a noisy background (SNR=-6dB) are used. Three sets of listening tests are pursued: (i) spatial percepts of distance, width, focus and envelopment are evaluated in quiet; (ii) the same percepts are evaluated in noisy conditions and (iii) speech intelligibility is measured in noisy conditions only. It is shown that the percept of distance does not differ significantly between quiet and diffuse noise, while width, focus and envelopment are significantly altered. The effect of noise on them largely depends on the nature of the early reflections. In some case diffuse noise restores spatial percepts that are unavailable in the reverberation-only (quiet) conditions. This finding is traced back to the unmasking of the early reflections which were masked by the reverberant tail. When speech intelligibility is measured, a modest improvement is obtained with scattered early reflections, but the direct association between the spatial percepts and speech intelligibility in diffuse noise is weak and limited only to focus.

Last modified 2022-01-24 16:11:02