Hall et al. measured the masking threshold for a sinusoidal signal in one of their experiments, in which the center frequency was 1 kHz and the duration was 400 ms, as a function of the bandwidth of a continuous noise masker, keeping the spectrum level constant [2]. They used two types of masker, a random noise masker and an amplitude modulated random noise masker, which were both centered at 1 kHz, as follows: The former had irregular fluctuations in amplitude, and the fluctuations in different frequency regions were independent. The later was a random noise that was modulated in amplitude at an irregular, show rate; a noise lowpass filtered at 50 Hz was used as a modulator. Therefore, fluctuations in the amplitude of the noise in different spectral regions were the same.
Fig. shows the results of that experiment. For the random noise (denoted by R), the signal threshold increased as the masker bandwidth increased up to about 100-200 Hz, and then remained constant. This is exactly as expected from the traditional model of masking. The auditory filter at this center frequency had a bandwidth of about 130 Hz. Hence, for noise bandwidths up to about 130 Hz, increasing the bandwidth the filter increased the noise passing through the filter, so the signal threshold increased. In contrast, increasing the bandwidth beyond 130 Hz did not increase the noise passing through the filter, so the threshold did not increase. The pattern for the modulated noise (denoted by M) was quite different. For noise bandwidths greater than 100 Hz, the signal threshold decreased as the bandwidth increased. This indicates that subjects could compare the outputs of different auditory filters to enhance signal detection. The fact that the decrease in threshold with increasing bandwidth only occurred with modulated noise indicates that fluctuations in the masker are critical and that the fluctuations need to be correlated across frequency bands. Hence, this phenomenon has been called ``co-modulation masking release (CMR).'' The amount of CMR in that experiment, defined as the difference in thresholds for random noise and modulated noise, was at most about 10 dB [2].