In this paper, a watermark technique for identifying friendly and foe is proposed during communication. The speech signal is processed in several stages. First, speech signal is partitioned into small time frames and the frames are transformed into frequency domain using DFT(Discrete Frequency Transform). The DFT coefficients are quantized and the watermark signal is embedded into the quantized DFT coefficients. At the destination channel quantization errors of received signal are regarded as the watermark signal. Identification of friendly and foe are done by correlating the detected watermark and the original watermark. As in most other watermark techniques, this method has a trade off between noise robustness and quality. However, this is solved by a partial quantization and a noise level dependent quantization step. Simulation results in the various noisy environments show that the proposed method is reliable for identification between friendly and foe.