Ultra low bit-rate speech coding: An overview and recent results

In narrow-band speech coding, specifically in the low and ultra low bit-rate ranges, a series of efficient quantization of the LP parameters using fixed-length as well as variable-length segment quantization (VLSQ) have resulted in a progressive reduction in the bit-rate from the 2400 bits/sec baseline of the LPC-10 coder down to 300 bits/sec and less. The VLSQ framework forms a generic basis of a class of segment vocoders within which various types of segments/units and unit-modeling have been explored, such as phones (in the phonetic vocoder), automatically derived units (phones and diphones), R/D optimal linear prediction, HMM based recognition-synthesis and unit-selection based paradigms. Recently, set within the original unit-selection framework, we proposed a joint spectral-residual quantization scheme which obviates the need for transmitting any side information about the residual of the input speech, offering up to 2dB spectral distortion at 250 bits/sec. In this paper, in order to realize better rate-distortion performance, we propose joint spectral-residual quantization in an optimal unit-selection framework based on a modified one-pass dynamic programming (DP) algorithm.