This paper presents a new microphone array method to enhance speech signals in a noisy reverberant environment. A time-delay estimation method is used for the speech source localization. The robustness of the localization method in high noise levels is provided by a subband Kurtosis-weighted structure. The estimated inter-sensor time-delays are directly used in an adaptive soft-constrained subband beamformer. Evaluation in a simulated environment with real speech sequences shows promising results.