[Speex-dev] Abuot the Bug of Speex 1.2beta1

王立传 wlcgyq at yahoo.com.cn
Thu Nov 23 20:59:13 PST 2006

Dear Sir/Miss:
    I'm very sorry to bother you,but I've found a bug
in the Speex,Version 1.2beta1,and also I've have some
    I'm deeply impressed by the work you have done in
the coding and decoding of speech signal.But when
using speex,I encoutered a bug,that is when in 32K
mode,the window size is still 200 samples,but not 400
samples,the bug will bring the problem of the decling
of the codeing quality,and the speed is very low.
   Also,I have some questions,and I need your help. 
   The fisrt question is that,I wonder why in
1.2beta1,the window is center flattered,rather just
like previous versions, use the simply the
asymmetrical window.
   The second question is more important and more
complicated,if possibal,plese give me a detailed
description.In useing speex,I've found that although
the coding quality is great,the speed is a bittle low,
so i want to use SSE instructions to improve it,but
when using vector quatilization,speex use
sub_vectors,and many of  the sub_vectors sizes are
like 5,10,but not the times of 4,so SSE instructions
can not improve so much.So I want to know,how do you
train and get the fixed codevect,and using your
method,can I train and get the codevect,and the
subvect size is 4,8,12,16,20,and coding quality is
almost the same?
   Thank you very much!
   Best Wishes!
                                        Grain Wang



More information about the Speex-dev mailing list