On the robustness of HMM and ANN speech recognition algorithms