Okay, after thinking about this, I am just too curious. This is just a discussion, not an attack.
What codec are you using?
What is your basis for saying you can tell the difference?
I will tell you my limitations. Years ago I did some abx testing and even back then for, statistically, for most people, including me, 128 kbps lame mp3 was pretty tricky to abx from the original. At 192 kbps any difference was to the point that for me abx testing was exasperating and boring and pointless, once I tried it a few times. For some demanding samples I could sort of tell the difference at 128 kbps, for others definitely not, and the differences were small enough that I would lose interest and patience. There were people who could do it systematically and reliably but in many cases by their account that was because they had taken the time to learn exactly what the compression artifacts sound like. They were of a nature that would skip most people’s attention, they said, and they could do it even with modest headphones. The quality of the gear made little difference in terms of being able to detect artifacts.
If I remember right even a 1db difference in volume is easily abx’ed.
So if most people are having trouble ABXing something perhaps you can see the differences are probably quite minor.
So the fact that even almost 10 years ago 128 kbps was sort of the borderline for transparency for a good mp3 codec makes me wonder about the basis of your claims. Saying that you can tell the difference between 128 kbps and the original is plausible to me but not probable, unless you are using an inferior codec. Saying you would not use anything less than 192 kbps for speech seems a little over the top to me. Since I originated this thread 8 1/2 years ago I am interested in where you are coming from. If you want to dig deep where people tried to do this stuff right you can hop on over to hydrogen audio and look through very old abx testing threads, though I think for them they beat the subject into the ground as the technology progressed until the codecs were too good to make it interesting anymore.