The traditional approach to reduce acoustic echoes is the so-called acoustic echo canceller (AEC) which subtracts an echo estimate from the microphone signal yielding a local near-end signal estimate. This method models the echo path with an adaptive filter. To obtain a satisfactory echo cancellation in a reverberating environment, an adaptive filter of high order is necessary which makes the AEC highly computationally complex. In contrast, the method proposed in this thesis which is called acoustic echo suppression (AES), follows a different approach where the echo reduction is performed with spectral modification. In order to perform a spectral modification in a non-uniform bandwidth subband domain, a M-channel GDFT-modulated filterbank is used where the delay elements are replaced by allpass filters of first order. Using a subband decomposition with non-uniform bandwidths, the critical bands of the human ear can be approximately matched. An alternative structure called filterbank equalizer (FBE) that only consists of analysis filterbank and a simple summation as the synthesis stage is introduced with the objective of reducing complexity and system delay. As weighting rule for the spectral modification a parametric Wiener filter is used. A set of adaptive filters are employed to compute an echo estimate in the subband domain. Both echo estimation in subband domain and weight computations are performed at a low sampling rate. To obtain a better performance during doubletalk situation, a doubletalk detector based on normalized cross-correlation is also implemented that calculates a decision variable per subband. For simplicity, only one of the decision variables is used in order to freeze the adaptation during doubletalk. The proposed approach is simulated and evaluated by measuring the estimation mean square error. Informal listening was performed to adjust the different parameters. In conclusion, the presented scheme has a low complexity and exhibits a satisfactory echo reduction.