15th Speech in Noise Workshop, 11-12 January 2024, Potsdam, Germany 15th Speech in Noise Workshop, 11-12 January 2024, Potsdam, Germany

How the types of sound reflections influence speech intelligibility in rooms

Nicola Prodi
Department of Engineering, University of Ferrara, Italy

Depending on the room boundaries, the sound reflections following the direct sound in a room can be either specular or at least partly diffusive, the latter also named scattered. While specular reflections are correlated by nature, scattered ones are usually uncorrelated. Still limited research has been directed to the understanding of how our auditory system incorporates the early sound energy from either type of reflection and the results are not entirely consistent. The amount of early sound energy is highly relevant in the context of speech intelligibility of a target source in rooms due to the time limit of the useful energy integration. In addition, part of the binaural information coded in the target impulse response can be used to estimate the binaural unmasking. For this scope the interaural phase difference between target and masker is used in E-C based models together with the masker correlation. In similar modelling E-C frameworks the correlation of the target signal is almost disregarded: this is realistic as far as short source-receiver distance is considered or if the environment is very damped. Actually, in a real context, target correlation may vary, and is modulated either by the type of reflection, by reverberation, by distance or by a mixture of all the factors. Unfortunately, the accurate control of target correlation alone while keeping other speech qualities untouched is not trivial. In recent years much effort was made to disentangle the role of target correlation from other known monaural and binaural factors influencing speech intelligibility. In one experiment based on simulations the whole impulse response consisted of either specular or scattered reflections while monaural and binaural indicators were kept fixed. The noise was localized and it was either energetic or informational. The obtained spatial release from masking (SRM) differed for the two types of reflections, specular and diffuse, and the former ensured better performance. The E-C modelling approach helped to isolate the contributions of better-ear and binaural unmasking. In another experiment based on real measures the target correlation was further investigated by comparing symmetric scattered early reflections (that, although diffuse, keep an high left-right correlation) against unsymmetric ones that provide very low left-right correlation. The interferer’s correlation was modulated too. Results helped in better understanding the role of target correlation and provide grounds for better acoustical design of rooms for speech.

Last modified 2024-01-16 10:49:05