14th Speech in Noise Workshop, 12-13 January 2023, Split, Croatia 14th Speech in Noise Workshop, 12-13 January 2023, Split, Croatia

P07Session 1 (Thursday 12 January 2023, 15:30-17:30)
Spatial release from masking in sound fields with specular or scattered reflections

Nicola Prodi, Chiara Visentin, Matteo Pellegatti
Università di Ferrara, Italy

Background: Speech intelligibility in rooms depends on the amount of reverberation, and on the distribution of sound energy in the impulse response between early and later reflections, the former being considered beneficial, the latter detrimental. In addition, the correlation of the masker at the ear of the listener provides higher or lower efficacy to the E-C process whereby the masker can be spatially disentangled by the target hence affecting spatial release from masking (SRM). Specular reflections are coherent by nature, whereas diffuse scattered reflections are related to a temporally smeared wavefront and are intrinsically incoherent.

Rationale: Specular and scattered reflections may differently influence the degree of correlation of both source signal and masker at the listener’s ears. They could therefore have a non-negligible influence on the SI level achieved for a given reverberation time and fixed energetic ratios. In particular this should be proved by measures of SRM; the relative importance of the phenomenon for the target source and for the interferer should be possibly assessed.

Methods: Binaural room impulse responses (BRIR) having either specular or diffuse reflections were simulated. The levels of reverberation time (T20mid = 0.45 [s]) and clarity (C50mid = 11.7 [dB]) were fixed to values of an inclusive room for speech. An anechoic case was added as reference. Source-receiver distance was set at 3m. SRM listening tests were accomplished by means of energetic and informational noise at 0° and 105° by using the BRIR, convolution and a multi-channel audio rendering system.

Results: The BRIRs from specular and diffuse reflections had cross-correlation values that both differed from unity and had gaps between them larger than the just noticeable differences. SRM data showed the effect of reflections type, with specular ones providing larger SRM values in most cases. The effect of correlation of reflections was larger for the target source than for the masker. The joint effect was not strictly additive but larger than the two effects on source and masker in isolation. The effect of the type of noise was also observed.

Conclusions: The nature of the reflections in the BRIR is a factor that builds up the speech intelligibility performance via correlation. In a single room the effects on target and masker are necessarily to be considered both, even at realistic distances. The results provide new evidence of the practical implications of BRIR correlation and aid at better setting the optimal conditions for speech in rooms.

Last modified 2023-01-06 23:41:06