Auditory masking refers to the fact that the threshold of audibility of a target signal is raised by the presence of another (masker) signal. In simultaneous (frequency) masking, where target and masker occur simultaneously, the amount of masking generally decreases with the frequency separation between the two signals. In temporal masking, where the target is presented either before the beginning or after the end of the masker, the amount of masking generally decreases with the temporal gap between the two signals. Both effects have been studied extensively in the literature, but they have almost always been studied separately. Very little is known about the interaction between these twe effects, i.e. the spread of masking across time-frequency space.
A basic experiment on the spread of time-frequency masking is performed within a collaboration project between Marseille and Vienna. The masked threshold for a target signal is determined at various time-frequency positions around a masker signal. Both the masker and target are Gaussian shaped tones, the masker being fixed at 4000 Hz, and the target placed at all combinations of five frequency and five time intervals relative to the masker. The reason for using Gaussian shaped tones is that these stimuli probably best decribe the basic auditory time-frequency window (Van Schijndel et al., J. Acoust. Soc. Am. 105 (1999) 3425-3435). In one special condition, where both the frequency and time interval between masker and target are zero, the experiment is equivalent to an intensity discrimination task. This is because adding two identical signals causes an increase in intensity. While this condition appears to be very important to model the results, its interpretation is still a matter of discussion.
The results may improve our understanding of the interaction of temporal and frequency masking effects. Furthermore, the obtained data are intended to serve as a basis for improvements of a masking filter model which identifies and removes perceptually irrelevant components from an arbitrary complex signal. Currently, that model considers only the properties of frequency masking. A future version of the model will apply a 2-D convolution kernel, which has to be defined based on the experimental results. The incorporation of the properties of time-frequency masking is supposed to improve the efficiency of the model.