Speech elements near masked threshold carry intelligibility only under binaural listening.

Evidence

The question of interest is whether the effect of eliminating spectro-temporal regions will follow a different course for collocated and spatially separated source configurations. Should a BMLD-like effect indeed act at the level of small spectro-temporal regions, then deletion of target speech from spectro-temporal regions with local SNR values between ‑15 and 0 dB (i.e. between dichotic and diotic masked thresholds, Hirsh ( 1948 )), should degrade speech intelligibility for separated, but not for collocated sources, since those speech components would be inaudible in a collocated configuration anyway. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1620

Interestingly, humans are capable of detecting the content of sound coming from different directions in the presence of distracting sources. The speech signal has been widely used as a the test signal in a the study of this capability, as it presents a natural signal to which human hearing has adapted. In the studies, a speech signal S with masking noise(s) N is presented to the listener. The level of the signal is measured for both binaural listening and monaural reference listening so as to produce the same intelligibility in both cases, and the difference in level between the monaural and binaural conditions is taken as the binaural intelligibility level difference (BILD) (Blauert, 1996). — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1513

Categorization

• DEFINE: Defines BILD (binaural intelligibility level difference) as the measurement metric for the binaural advantage.
• XREF: Relates directly to the cocktail party effect and spatial release from masking, a core acoustic-communication application.
• corroborated: 2 independent passages support this claim — Dijk, Pim van, Physiology, Psychoacoustics and…, 1620; Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1513