Gabor Multipliers Applied to Masking, from Theory to Application!

We have talked about an overview and a summary on the mathematical and signal-processing background for time-frequency masking filters based on the concept of 'frame multipliers'. Gabor multipliers are a current topic of research, known in signal processing as Gabor filters or time-frequency masks. They have several applications in computational auditory scene analysis, audio signal separation and speech recognition. Frame multipliers are a generalization of this type of time variant filters to frames without further structure. After analysis, before synthesis the coefficients are multiplied by a fixed pattern, the so called symbol. The dependency of the operator on the symbol and frames was presented here.

Masking filter algorithms are for example used in the MP3 coding, and currently are used in a number of applications. Their primary task is to filter signal components, which cannot be perceived by the human auditory system. This is certainly a non-trivial task. It strongly depends on the signal itself, and so it can be seen as an adaptive filtering, which is highly non-linear. But this filtering can be separated into two steps, first the calculation of the operator, which then is applied to the signal.  The second part is linear again. In the case of masking this means that first the mask for the time-frequency coefficients is calculated, which then is applied as a Gabor multiplier. A concept was presented how to implement a filter, which approximates both the simultaneous and temporal masking known in psychoacoustics with a numerical efficient algorithm.