Background: When computing the loudness of a target signal in the presence of other sounds, current computational models of loudness assume that the target signal and background noise are aggregated together on the basilar membrane without considering aspects of spatial hearing like the cocktail-party effect. In speech intelligibility, this effect has been shown to improve speech reception thresholds by up to 7 dB (Hawley et al., 2004, doi:10.1121/1.1639908). The present study investigates the partial loudness of speech in competing speech.

Method: The level difference necessary for equal loudness (LDEL) between the target in silence and the target in competing speech was studied for four signal-to-noise-ratios (SNRs; -6, -3, 0 and 3 dB). The target was always female speech at 45 degrees, the masker either a single male masker at the opposite side, or four simultaneous male speakers at 90 and 22.5 degrees on the same side, frontal and 45 degrees on the other side. Stimuli were recorded using an artificial head and presented via headphones. Twenty normal-hearing participants took part. LDELs were also calculated by a combination of the models of Glasberg and Moore (2005, 2005, J. Aud. Eng. Soc. 53:906) and Moore et al. (2018, doi:10.1121/1.5027246) that considers binaural hearing but not further processes of spatial hearing.

Results: The model predicted a decrease in partial loudness with increasing noise levels, reaching LDELs of -4 dB for the single-talker masker and -10 dB for the babble masker at an SNR of -6 dB. The experimental results followed this trend for the babble masker, reaching and LDEL of -7 dB at -6 dB SNR. For the single competing talker, all LDELs were close to 0 dB. A repeated-measures ANOVA confirmed these effects.

Conclusion: Overall, this study provided evidence for the importance of considering the effect of spatial separation in loudness perception. The surprising result of basically the same loudness in silence and that in the presence of a single competing talker that has 6 dB higher level suggests that partial loudness was probably judged during dips of the masker, and participants were able to fully focus on the target. The model predictions were reasonably close to the results for the babble noise.

Acknowledgments: This study was supported by the Royal Society, grant 232164.