Recent applications in integrated network and secure communication necessitate the development of flexible speech, coding techniques under adverse channel conditions. In this paper, we explore the vectorsum innovation codebook structure in conjunction with dynamic bit allocation strategy to develope an efficient variable rate speech coding system. Design algorithms for optimizing the basic innovation codevectors are described first. We then propose a dynamic bit allocation using the threshold decision rule to emphasize the locally important phonetic features. Furthermore, the linear prediction parameters are efficiently quantized by exploiting the inherent ordering relationship among the line spectrum pair parameters.