Single channel speech enhancement by frequency domain constrained optimization and temporal masking