Data hiding embeds extra information into digital media for the purpose of authentication, annotation and copyright protection. Transparency and robustness are two contradictory requirements for any data-hiding scheme.This paper studies audio data hiding using spread spectrum (SS) technique in subband domain. On transparency part, we propose a novel method of tuning psycho-acoustic models employed in audio compression to control the audibility of introduced distortion for data hiding purpose. On robustness part, we have studied feature selection and synchronization problem in order to maximize the survivability of the embedded data. Our study shows that mid-band coefficients are appropriate features for data embedding. In addition, we propose to track/recover synchronization of audio signal before the detection, facilitating the extraction of the embedded data. Experiment results have shown that our data-hiding scheme in subband domain can survive a wide range of attacks while providing transparent audio quality and abundant embedding capacity (>20bps).