A well known technique for speech enhancementis spectral subtraction. This technique is attractivedue to its simplicity and its low computational cost.We have proposed a simple modification of the filtercoeficient calculation in form of a recursion of theprevious filter coeficient. Because of this recursionthe subtraction rule (gain curve) consists of twodifferent paths, one for increasing signal-to-noise-ratio and another for the decreasing ratio. In thispaper we will introduce a parametric version of thisrecursive approach with two parameters controllingthe crossover behaviour of the two paths. Althoughthis recursive version is a very simple algorithm itis easily adjusted to produce no musical noise andadditionally to get low speech distortion and/or lowresidual noise.