Frequency domain speech coding techniques such as sub-band coder (SBC) (ref 1) and adaptive transform, coder (ATC) (ref 2) can produce high quality digital speech at around 16 Kb/s. However, at bit rates below 16 Kb/s, their rapidly deteriorating speech quality and increasing complexity make them less competitive to the time domain coders such as residual excited linear prediction (RELP) (ref 3). In this study a hybrid of ideas from ATC, RELP, and vector quantization (VQ) are put together in order to improve the quality of a low bit rate transform coder. Informal listening tests have shown that the proposed coder out-performs the speech specific ATC (SSATC) (ref 2) at 9.6 Kb/s as well as SBC (ref 1) and RELP (ref 3).