Coding of speech signals.

The invention relates to a method for encoding speech in a speech signal sampled using a long-term prediction (LTP). DETERMINED DELAY parameter STEP FOR EACH PAGE LTP speech signal, first determining the autocorrelation function for the page within the signal, between about DELAYS MAXIMUM AND MINIMUM DEFAULT. THEN weighted autocorrelation function to emphasize the function, for existing DELAYS IN THE VICINITY OF delay parameter STEP THAT HAVE BEEN DETERMINED TO PAGE NEWEST VOICE. It is then VALUE FOR MAXIMUM WEIGHTED autocorrelation function and identified as the delay parameter PASS FOR PAGE.