In December 2005 the PerenniAL Acoustic Observatory in the Antarctic Ocean (PALAOA, Hawaiian whale) was set up on the Ekström ice shelf, Antarctica near the German Neumayer Station (Boebel et al., 2006). Since then it almost continously records the underwater soundscape in the vicinity of the shelf ice edge and is intended to do so over the duration of several years. These long-term recordings allow studying the acoustic repertoire of whales and seals in an environment almost undisturbed by humans. The data is analyzed to detect species specific vocalizations, infer the approximate number of animals inside the measuring range, calculate their movements relative to the observatory, and examine possible effects of the sporadic shipping traffic on the acoustic and locomotive behaviour of marine mammals. Underwater sound is recorded by means of four hydrophones located through boreholes below the 100m thick floating ice sheet. They are attached to an autonomous, wind and solar powered station, which can record at 192kHz/24Bit. A compressed data stream is transmitted in real time via wireless LAN from PALAOA to the German Neumayer Base at 15km distance. From there, a permanent satellite link transmits an even more compressed stream to the AWI in Germany. It can be accessed live from our webpage at http://www.awi.de/acoustics. So far, Weddell seals, crabeater seals, Ross seals, leopard seals, killer whales, blue whales, fin whales and minke whales have been identified in the recordings along with several vocalizations which could not be assigned certainly to a species yet. Additionally many non-biological sounds were recorded, mostly generated by ice and some anthropogenic events like ships passing by and human activities on the ice. Difficulties arise from the sheer size of this constantly growing dataset, more than 7500 hours so far. We develop interfaces and setups to process this stream in real time and analyze it both interactively and by means of batch processes running in parallel on a workstation cluster, for example applying detectors specific to single species, based on hidden Markov models. These recordings, which are largely free of anthropogenic noise, provide also a base to set up passive acoustic mitigation systems used on research vessels by developing automatic pattern recognition procedures to be used in the presence of interfering sounds, e.g. propeller noise.