Efficent quantization of synthesis filter coefficients on CELP (Code Excited Linear Prediction) coders is very important for a good performance of this class of coders, specially when operating below 8 Kbps. Three vector quantizers with good potential for utilization on low rate coders are studied. Each of them was implemented inside the structure of VSELP( Vector Sum Linear Prediction) coder, an important menber in the class of CELP coders: VSELP is the north American cellular Communication standard. These three vector quantizers work with LSP parameters and are compared considering the robustness to channel errors, complexity and quality of sythesized speech. The quality of synthesized speech is evaluated considering objective measures of frequency weighted signal to noise ratio and subjective results obtained from informal listening tests. In order to improve the robustness to channel errors, the problem of binary indices assignment to the output levels of the quantizer is also investigated. The technique used to determine good indices assignments is Simulated Annealing