When I try the "Ephraim-Malah Noise Suppressor" example (speech-recAudio-ProcessingNR_exmp) with IPP 5.0 or 5.1, I get different results on different processors.
After hours of debugging I found the some implementations of the ippsUpdateNoisePSDMCRA_32s_I routine don't work correctly.
When I try the generic (px) and Pentium III (a6) versions of this routine, the noise cancelled audio contains drops in amplitude, which is disturbing and distracting.
When I use the SSE2 (w7) and SSE3 (t7) versions, this issue doesn't occur and the result is much better.
Is this a known issue?
I need at least a generic version of the routine that behaves like the SSE2 or SSE3 version.
Thanks for any help!
