Neutral, transparent, warm

I would listen to a lot of speakers and then chose the one that sounds best to you.  It also depends upon the music you listen.  If you like to hear crisp highs like cymbals then choose tweeters that can produce this sound.  If you have some local sound shops listen and talk to them.  If you don't have good local stereo retailers then jump in the car and visit some.  Sound also depends upon the amplifier driving them.  Each has its own sound.  Try to not make a quick decision.  I would listen to the ones you like best for about two hours.  Bear in mind, when listening can you imagine yourself listening to them for hours or do they either lack something or do they cause hearing fatigue.  Sound should be relaxing to your ears, as you are listening to things you like to hear.

If you like to hear crisp highs like cymbals then choose tweeters that can produce this sound. 

Or if the speakers do piano and the human voice, then is also a good sign that they will do everything else well.

bdp24, Nice post explaining terms in easy to understand ways.

These are simply a few opinions after hearing the brands over the years:

Sonus Faber, usually warmer, good resolving power.

B&Ws, Focal, Dynaudio brighter with good resolution.

Golden Ear, more neutral, good resolution.

  Your ears may not agree. Me, I love transparency that is not harsh but rich with tonality. I'd lean to Sonus Faber these days, not B&W (although I own a little pair).

Well piano and the human voice are difficult, to be sure. Harbeth fans always describe how the Harbeth does voice so well and to my ears voices sound colored and too nuanced through these speakers. Having said this I would rather listen to a pair of Harbeths than many of the B & Ws or Focals that I have heard in the past, especially if the software is a bit on the rough sounding side. 

One more observation:

JGH was particularly bothered by and critical of a particular type of coloration, one very common amongst loudspeakers. He even came up with a term for it: "Vowel coloration." He defined and explained the term in one early-70’s Stereophile speaker review I read, and I immediately understood of what he was speaking.

LIstening as much as I do to music containing harmony singing, my first test of a loudspeaker is in its’ ability to reproduce voices "accurately". Hearing voices as much as we do, it’s easy to hear when a speaker is adding coloration to singing voices; most do, to one degree or another. When I hear vowel coloration being added to voices, the speaker doing so is immediately crossed off my list of acceptable choices. I have established a very high bar in that regard.

In 1973-4 I made recordings of my wife and young son speaking, using a nice condenser microphone JGH had reviewed in Stereophile plugged straight into a Revox A77. I also recorded (with a pair of the mics) the Jump Blues/Swing band I was then playing in, both in a rehearsal space and live in a local bar. Upright piano, electric bass and guitar, tenor and baritone saxes, my Gretsch drumset and Paiste 602 cymbals, and male vocals.

Both recordings serve as excellent references when evaluating loudspeakers, better than almost all commercial recordings, most of which have been subjected---to one degree or another---to all kinds of electronic manipulation, rendering the "accuracy" of the recorded sound unknowable. What should they sound like? Who knows?!

One notable exception are the direct-to-disc LP’s on the Sheffield Labs label. Stunning lifelike sound: extremely transparent, with incredible immediacy, dynamics, and lifelike instrumental timbres. Get yourself some, and use them for your loudspeaker evaluations. If a speaker is adding coloration, reducing transparency, or both, you will immediately hear it.