This paper presented Compressed Speaker Recognition (CSR), which is a scalable approach for performing speaker recognition directly from live, compressed VolP packet streams CSR creates a discriminating feature vector directly from the compressed VoIP packet stream. This eliminates much of the time consuming processing required by traditional approaches based on the decompress-FFT-Me1-Scale filter cosine transform. CSR employs a high-speed micro-clustering technique which allows it to analyze much higher bandwidth rates than any existing system. The accuracy of CSR has been examined and shown to be compatible with that reported in the NIST speech group and evaluation report. At the same time, CSR requires less time for the recognition. Importantly, CSR demonstrates good scalability in terms of CPU utilization and memory requirements. The CSR approach analyzes directly from the compressed voice packets This will allow the compressed voice data packets to be sub-sampled and reduce the memory buffer requirement This enhances scalability even further.