A novel non-blind digital audio watermarking scheme using time-frequency based operations is presented. The digital audio samples are first segmented into frames and the frames are transformed into the frequency domain. Frequency components are then added or removed. The amplitude of the components added or removed are determined under the guidance of Psychoacoustic Model I. Inverse transform is then performed to restore it back to a time domain signal which is a lengthened (expanded) or shortened (compressed) version of the original. The expanded or compressed frames are then used to code the watermark to be embedded. Informal listening reveals that there is no perceptible distortion of the audio signal for suitably compressed or expanded segments. Recovery of watermark is perfect for signals not subjected to attacks. Experimental results also show a high degree of successful recovery of the signature that allows conclusive authentication of the owner even after filtering, cropping, MPEG compression, resampling and the addition of noise.