Visual information can help listeners to better understand what is said. In the speech coding domain, it will be shown that it allows to reduce the transmission rate of a classic vocoder (1,9 kbit/s instead of 2,4 kbit/s) by estimating audio parameters from video ones. In addition, vector quantization seems to be a good method to reduce the redundancy between some audio and visual coefficients. With the vector quantization, we can reduce again the bit rate while decreasing the quantization error.