FIG. 6.

(Color online) The modulation frequencies that contribute to the overall ENVneural metric, which predicts intelligibility, depend strongly on the envelope spectrum of the masker. The target PLV spectra shown are z-scored with respect to a null distribution that is common across conditions. The dashed lines indicate z = 1.64, i.e., the 95th percentile of the noise floor distribution. Comparing the modulation spectrum of speech-shaped stationary noise (rightmost panel) to the target PLV spectrum for the –2 dB SNR SiSSN condition (A), we find that speech-shaped stationary noise degrades the representation of high-frequency target modulations more (and low-frequency modulations less), in line with stationary noise containing relatively more high-frequency modulation power. In contrast, comparing the modulation spectrum of four-talker babble (rightmost panel) to the target PLV spectrum for the 4 dB SNR SiB condition (B), we show that four-talker babble degrades the representation of low-frequency target modulations more (and high-frequency modulations less), consistent with babble containing relatively more low-frequency modulation power. These results show that the spectral profile of EEG-based target-envelope coding fidelity (i.e., the neural envelope-domain SNR in target-speech encoding) is shaped by the masker's modulation spectrum. This result, in combination with our finding that EEG-based target-envelope coding predicts intelligibility, provides further neurophysiological evidence for perceptual modulation masking. These results also suggest that the modulation frequencies that contribute most to speech intelligibility in everyday listening could lie anywhere in the full continuum from slow prosodic fluctuations to fast pitch-range fluctuations.