Use of Pitch Prediction and UN Classification in CELP Coders
R. Ahn, Dipanjan Sen, W.H. Holmes · International Symposium on Information Theory and its Applications · 1994
There have been many attempts to reduce bit rates in voice coders based on the Code Excited Linear Prediction (CELP) method. One area of the CELP coder that is often overlooked by researchers is the adaptive code book analysis. The exhaustive search method which is currently used in most CELP coders is not only computationally expensive but also often produces unwanted pitch doubling and halving within voiced regions. This paper will present a method of CELP analysis in which a pitch detector is used to search for the optimum pitch delay in the adaptive code book. Furthermore, by using the unvoiced/voiced information, which is obtainable as a by-product of the pitch detection algorithm (PDA), an efficient way to use available bits to improve quality is proposed. This new CELP coder employs different combinations of adaptive code book size and stochastic code book size in voiced and unvoiced sub-frames. Informal subjective listening tests show improvement in synthesised speech quality when compared to the FS-1016 CELP implementation.