An audio watermarking scheme STAW in short time discrete cosine transform domain using spread spectrum method is proposed in this work. Here, we framed the audio signal in the time domain with overlapping segments which then transferred as DCT represented, according to binary watermark, we generated some pseudo-random float sequences as embedding message, which are orthonormal and energy uniform, since the DCT coefficients is not integer or binary. Additionally, in order to improve imperception an adaptive mechanism for embedding strength is proposed. The experimental results show that STAW provides high robustness against common audio attack as well as high imperceptibility.