Dear Speex developers, I think I found a typo in libspeex/preprocess.c (that still in the SVN version too). Please correct me if I would be wrong. Patch is attached. (Yours :parenthetically &dezzy) -------------- next part -------------- A non-text attachment was scrubbed... Name: noise_var.patch Type: text/x-patch Size: 518 bytes Desc: not available Url : http://lists.xiph.org/pipermail/speex-dev/attachments/20060805/584ca4d2/noise_var.bin
Thanks. Fixed in svn now. BTW, does it actually improve things for you? Jean-Marc Le samedi 05 ao?t 2006 ? 14:03 +0200, dezzy a ?crit :> Dear Speex developers, > > > I think I found a typo in libspeex/preprocess.c (that still in the SVN > version too). Please correct me if I would be wrong. Patch is attached. > > > (Yours :parenthetically > &dezzy) > > _______________________________________________ > Speex-dev mailing list > Speex-dev@xiph.org > http://lists.xiph.org/mailman/listinfo/speex-dev
On Sat, 2006-08-05 at 22:25 +1000, Jean-Marc Valin wrote:> Thanks. Fixed in svn now. BTW, does it actually improve things for you? >Nope. I tested the different versions ("typoed" and "untypoed") on the same samples, but I did not notice any significant improvement. Although, I have a question about the mechanisms implemented in Speex. I started to study the (narrowband) CELP [nb_celp.c] implementation, and I found the whole frame analysis part ("open-loop estimation of pitch and excitation gain") a bit unexplained: it computes LPC values and then transform them into LSP values, but why does it need to enforce margins between those LSPs? Does it have any special reason? Thanks in advance. d.