The paper presents a method of  LSF vector quantization of speech signal 
with forecasting the next value. The main idea of this method is thatit 
is encoded the difference between real and predicted values of LSF 
vectors instead of quantization of a real vector. 
This approach reduces the dynamic range of input values and thus it reduces 
the quantization error. The iterative codebook designing procedure 
is developed for the proposed method. 
The experimental results of the method testing are presented for different 
encoding velocities of a speech signal.