|
Hi all! I have another interesting query on ACELP wrt G.723.1 (well, atleast i think its interseting!) As far as ACELP goes, the search is performed in 4 nested loops, where in each loop the contribution of a new pulse is added. The objective is to maxime the correlation C. Now, the crux of ACELP is "focused search approach". Here a precomputed threshold is tested before entering the last loop and the last loop is entered only if this threshold is exceeded ie (absolute correlation > threshold). What is the logic behind this threshold being a "lower" bound? what does it mean if (threshold < absolute correlation). let us say that there is a heuristic upper limit to the contribution due to the first 3 pulses.beyond this the contribution of the 4th pulse wouldn't matter much. So if the contribution due to the first 3 pulses is greater than some threshold, we need not search the 4th pulse exhaustively...perhaps skip the last loop. by this logic, 4th loop should NOT be entered if threshold is exceeded. but the recommendation does exactly the opposite. Can anyone throw some light on this issue? even suggestion of relevant literature/paper would be great. You see, the very essence of ACELP search is the above simplification...and i am unable to figure it out! Thanks a lot Sameer Kibey Pune,India. |