The way real time pitch shifting works is the input is split up into "granules" which are little snippets of audio. Those granules are played back either faster or slower to shift the pitch up or down respectively. An analog analogy is a tape recorder with rotating heads. As these heads rotate they eventually leave the tape and the other head enters. As this happens the heads crossfade. If the audio isn't perfectly correlated during the crossfade the amplitude will vary. This causes the tremolo. The pitch detector's job is to find the pitch so that the audio is correlated during the crossfade.
If your intonation is off the correlation will be weak during the crossfade causing more tremolo.
Pitch shifting isn't perfect, especially in real time. It's a perceptual process. It's not like solving equations or modeling amplifiers. It's about fooling the human auditory system. So any pitch shifter is going to have some level of artifacts. In my comparisons with other products I feel these latest algorithms are among the best. I compared my Eventide Eclipse, Kemper and Digitech Whammy DT.
The Eclipse is the weakest of the bunch. It tracks very slow which causes artifacts when you change the note you are playing. It does, however, track complex chords well. For example, if you are playing a harmony the harmony will change from, say, a major 3rd to a minor 3rd depending upon the note you are playing. The slow tracking causes the shifter to sound the wrong interval for a bit until the tracking catches up. So you get a pitch "blip" when you change notes. It has a pleasant characteristic sound which has an almost bell-like quality that makes it useful for certain special effects.
The Kemper has pretty good shifting and tracks much faster than the Eclipse. It exhibits the most tremolo of all the devices tested (I know why this happens but I'm not about to spill the beans
) It struggles, however, with chords. If you play a D/F# (the classic pitch shifter stress test chord) it goes all wonky because it can't figure out the pitch. This is because it's pitch detector window is too short. It has a "smooth chords" mode which doubles the window length but then the latency becomes much greater and the tracking much slower. It also has a formant shifting feature but in my tests it's all but useless. It only works with single notes, adds a lot of latency and doesn't really sound convincing.
The Whammy DT is an interesting device. It exhibits almost no tremolo effect regardless of the chord played. The Eclipse, Kemper and Axe-Fx all have similar sounding shifting. IOW they seem to be based around similar algorithms implemented to varying degrees of success. The Whammy has a fundamentally different sound. This is both good and bad. The lack of tremolo is good. However when playing even simple chords (i.e. an open E or A) there is an underlying grittiness and "inharmonicity" and the guitar sounds out of tune. It also exhibits a strange behavior during the pick attack. The Whammy glitches during the pick attack kind of "cuts out".
For the Axe-Fx I'm using a wavelet approach to the pitch detector which results in a multi-resolution pitch detection. This means the pitch detection window is short when you want it short or long when you want it long. This results in extremely fast tracking. There's almost no audible "blip" when the harmony changes. It also tracks complex chords pretty well but it does sometimes struggle with locking onto the pitch and will alternate between the "best guess" when there isn't a clear answer. To be fair the other products do this too.
So all the devices exhibit their idiosyncrasies and artifacts. The question becomes which artifacts are the least offensive. If I were to choose I would choose either the Whammy or the Axe-Fx. However the Whammy only supports chromatic shifting. It has a "Harmony" mode but it isn't diatonic harmony where you can set the key and scale/mode.