|
Hi everyone: I am working on the quantization of LSFs in G.723.1. I found that sometimes G.723.1 introduces a quite large percentage of outliers in the range of 2-4dB, for example, over 10%. So G.723.1 does not satisfy the transparent quantization? Is this correct? Could anyone tell me where can I find the LSFs' quantization performance of G.723.1? |
|
|
Quantization performance of LSFs in G.723.1
Started by ●May 5, 2004
Reply by ●May 10, 20042004-05-10
Hi,
Yahoo! India Matrimony: Find your partner online. |
|
|
Reply by ●May 11, 20042004-05-11
|
CELP ACELP & RCELP is all based on LPC.
LPC means Linear Predictive Coding;
CELP means code exited Linear Predition ;
RELP means Residual Excited Linear Prediction;
ACELP means Algebraic Code Excited Linear Prediction,which is a simple
version of CELP.
You can get the detailed imformation by using Google searching
engine.B James <s...@yahoo.co.in> wrote:
|
Reply by ●May 16, 20042004-05-16
|
(Sorry, this messsage can appear twice in this newsgroup because of Yahoo). CELP is the name for the family of the speech coders based on analysis-by-synthesis (examples - G723 5.3 Kbps and 6.3 Kbps, FS 1016, G729 etc). ACELP - is a coder from CELP-family with algebraic codebook (example - G723 5.3 Kbps). MELP - belongs to the family of parametric coders (STC, WI, MELP, FS 1015), and it is significantly different from CELP although it also uses LPC. About zero excitation. Filter is linear operator WITH HISTORY. The history is what makes the output of filtering zero excitation non-zero. Symbolically it can be presented as follows. Suppose A is the filter with the history, A0 is the same filter with zero history, x is the signal to be filtered and 0 is zero signal. Then you can easily check that the following relation holds: A*x = A*(x+0) = A*0+A0*x Both elements on the right side are generally non-zero! If you understand this formula you have got the key to understanding the analysis-by-synthesis method. Ilya Druker |
