We mean the "human" speaker here, not the loudspeaker. Recording studios call them "voice talents". This is probably the most impalpable parameter of all, but it is also one with a huge influence on the ultimate result.
Experience shows that just as there are "radio voices", there is something like a "telephony voice." Some people have the ability to make themselves more clearly understood over one medium than over the other. A good radio voice doesn't necessarily make a good telephony voice and vice versa.
Most telephony cards work at 6 KHz or 8 KHz sample rates. This means that the highest voice frequency component these cards can actually play back to the phone line never exceeds 3 or 4 KHz. Worse, the telephone system itself limits the bandwidth to 3.4 KHz. It is therefore usually better not to select a voice talent with a lot of high-frequency content: a lot of what they pronounce would get lost in the process anyway. This is detrimental to intelligibility.