The local cache is a place where temporary copies of objects are kept. With a local cache in operation subsequent requests for previously cached object in the cached copy of the audio will be returned to the user. Since the local cache creates little or no extra network traffic, it is a good strategy to utilize it in streaming to improve the quality of streaming audios, especially the scalable audios. In this paper, we present an innovative cached-based streaming system for MPEG-4 scalable to Loss less (SLS) audio. The proposed system achieves better utilization of cache memory, and optimal playback quality under the conditions of limited network bandwidth resource by exploiting the scalable property of SLS audio. The SLS audio being cached are either the full-quality SLS audio files or only data units (DUs) from their base quality layers depending on how frequent they are expected to be played by the end user. The hit ratios of the cached contents are dynamically predicted by the Kalman filter in this paper.