Conditions and Results

We will consider as experimental stimuli ten pairs of two types of mixed signals used in previous section, f_M(t) and f_R(t). Although the simulation for CMR [19] was carried out for a function of the bandwidth of noise, this simulation is carried out for a function of a number of adjacent auditory filters L related to the bandwidth of a bandpassed noise. The amplitude envelope $\hat{\hat{B}}_k(t)$ in physical constraint 3 is determined by

$$\hat{\hat{B}}_k(t)=\frac{1}{2L}\sum_{\ell=-L,\ell \not= 0}^L \hat{B}_{k+\ell}(t)$$ (31)

while the input phase can be uniquely determined from Eqs. () and ().

Relationship between the bandwidth and the improved SNR is shown in Fig. . In this figure, the vertical axis shows inversely the improved SNR of a sinusoidal signal and the horizontal axis shows the bandwidth in relation to L. The real line and the error-bar show mean and standard deviation of the SNR, respectively. It was shown that, for the mixed signal f_M(t), the SNR of sinusoidal signal $\hat{f}_1(t)$ can be improved as the number of adjacent auditory filters L increase. In contrast, it is shown that, for the mixed signal f_R(t), $\hat{f}_1(t)$ cannot be improved as L increases. Here, improvement of the extracted sinusoidal signal is equivalent to masking release. These results show that the masking of a sinusoidal signal can be released as a function of the bandwidth of bandpassed noise when noise is AM banpassed noise, and that it cannot be released as a function of the bandwidth when noise is bandpassed random noise. For reasons, the proposed model can be interpreted as a computational model of CMR.