The difference is that there are 2 types of latency in play here - first, a latency in the system that is based on processing time between when the MIDI event is received and the corresponding audio is produced from the synth, and second, a latency that is perceptual based on the 'attack' of the sample, ie. a delay between when a sample starts to play and when it is effectively heard.
 
The first type of latency can (and usually is) automatically adjusted during playback. The DAW and plugin can work out between them how long it takes for a MIDI note to trigger audio and can read the MIDI early to compensate. (Or can push the audio back, but the end result is the same.) The idea is that sample 1 from the soft synth ends up playing back on exactly the frame when the MIDI On event is triggered. This doesn't help during tracking however because that latency is part of the system and (as we've noted) the DAW is incapable of predicting when you are going to press a key on your keyboard. So you will probably hear more latency while tracking than when you play it back.
 
The second type of latency is inherent to the sample. If the sample has an attack that takes 50ms before the main transient or peak is heard, it will feel like the sound has a 50ms latency relative to the MIDI event that triggers it. This is going to be here during tracking but it's also not going to go away during playback. Even with the usual latency compensation, a MIDI event at 1.000 seconds is going to result in that transient/peak being heard at 1.050s because that's how long it takes for the sound to 'get going'. If you know your instrument has that kind of slow attack, you can adjust for that with the Time+ option so that the MIDI is triggered slightly earlier to counterbalance it. But it can't be addressed by plugin delay compensation in the normal sense because the plugin's delay is already compensated for. Sample 1 is already playing exactly when the MIDI On event happens - it's just that you don't hear much until Sample 2000 or so.
 
Now, the plugin could 'lie' to Sonar and pretend it has more inherent latency than it really has, meaning that playback would be more likely to sound like it coincided with the MIDI. But that wouldn't work well if there are a variety of different attack durations (quite likely if some are plucks and some are strums), and either way it would make no difference during tracking because that compensation can't occur in real-time.

Its the first type thats the issues here though isn't it. As with the second the comparison is being made against 'like' instruments, so the attack is the same. The first is the response of the instrument, and fro I can make out, complaits are that many other VTis have a lower latency than Amplesound.
 
I'm not bothered by their guitar stuff, but the Bass stuff (J-Bass) was firm on my list until I read this thread. Playing a bass synth as part of a rhythm has got to be tight, and programming lacks feel.

The makers are on record as saying the 50ms delay is part of the sound ("it takes around 50ms to get the string really vibrating. We preserve the moment of the stroke, otherwise it will sound like piano.") rather than a processing delay.
 
If I was them I'd try and implement a zero-attack version suitable for tracking (eg. start the samples right before the transient with a very short fade-in) and use the full sample only on playback. I've never written a VSTi though so I'm unsure how practical that is.

Wow, is that really the response time of a physical guitar. Doesn't the Haas effect says that a 30ms delay causes a sound to be perceived as a separate sound than another How is this not an issue for a actual guitar player. Why does a piano string respond so differently?
 
I'm learning guitar, and just at the open chords sort of stage of playing, so complete beginner. I'm not interested in their Guitar stuff, because I think I should make playing the real one the aim. I have both strum acoustic and electric plugs, so if I wanted a substitute for composition I could use them. But then, being they are modelled on a physical string, isn't that an issue for them too?
 
Bass guitar strings must act similarly to normal guitar, so how does a bass guitar player keep tight, if it takes 50ms to reach its transient peak.

Scoot, I use Native Instruments Scarbee J-Bass (and Rick and P-Bass) and I have zero latency problems so maybe that's worth looking at instead of Ample Sound.
 
Note that I always set my Asio buffer low while tracking and high while mixing.
 

Cheers Konradh. Had a listen to the demos, I wasn't as sucked into the sound as the amplesound, but maybe that's down the the performances and I can set up more how I liked. It's a fraction of the price of the Amplesound though, and affordable gamble.
 
Another part of the appeal is that the Amplesound is dedicated to the one job. I considered Trillion, but it looked massive. Scarbee requires Kontact, which comes with a bundle of samples ( I know I don't have to install these, but I'd be intrigued). I'm trying to take a different approach and reduce my VSTi options. I've just installed Sonar on my new laptop, but I haven't installed any Sonar VSTIs. Just AAS full package, XLN Addictive Keys and Addictive Drums.

Scoot
Wow, is that really the response time of a physical guitar. Doesn't the Haas effect says that a 30ms delay causes a sound to be perceived as a separate sound than another How is this not an issue for a actual guitar player. Why does a piano string respond so differently?

I just did some unscientific studies by recording myself playing guitar, then zooming in on the waveform.
 
This is a single plucked note.

 
The first 20 milliseconds are partly pick noise and partly the string not resonating properly yet, probably due to it being distorted (by the pick movement) near one end of its length and time being needed for the vibration to travel along the whole string and start it all vibrating in unison. Between 20 and 30ms you see the waveform emerging, at 30ms it starts to get a lot louder, but it doesn't peak here until 55ms.
 
You could probably remove the first 20ms and nothing of value would be lost - but that still leaves a 35ms minimum delay to get a realistic-sounding pluck of one string.
 
You'll see the period here for each oscillation is roughly somewhere around 3.3ms and 3.2ms, ie. 303Hz. This is what you'd expect, because I've just plucked an open D which is 293Hz, but a plucked string always starts sharp and settles later.
 
Now, for conjecture:

• I suspect the Haas effect only matters for hearing 2 separate sounds. I don't think a gap of 30ms between motion and hearing will matter (otherwise, standing more than 30ft from a musician or speaker would mean you don't think they're playing or talking, as sound only travels at about 1ft per second).
• Not sure about piano strings, but since they are essentially percussion instruments I expect there is a slightly quicker attack. I also don't think it matters much for the reason above. I don't have a real piano to test with, so any test I could do will be on samples which may have already been trimmed. (I did take a look at NI's The Giant, and that did seem to have removed all the hammer noise so you could hear the fundamental almost immediately. I think TruePianos had a more realistic attack but I don't have time to measure it right now.)

 
 

I think its great that keyboard players have guitar VSTis now.  I wonder if any of them sound authentic?
 
I got the Scarbee Bass VSTs in Komplete, they sound good but programming them or playing them with the needed articulations is almost impossible, just easier to play a real bass.

Thanks fr taking the time to test the waveform, I had contemplated doing the myself, but am juggling a few things until after the weekend.
 
That initial sound is quite low in amplitude, so I agree you could probably dismiss it. Is it the sound of the pick releasing the string, or the initial press against the string. I suspect the latter. Although the peak is at 55 ms, the amplitude is substantial at around 44ms, so removing that 20 seconds pick press m leaves around 24 ms, even less if they treat treat the initial 30 ms as insignificant. The natural envelope of the sound is still there, so hard for them to say it suddenly sounds like a guitar.
 
I guess I'm just annoyed at their answer, because I had spent time looking at the manuals, checking reviews, and the demos and tutorials. It seemed just what I wanted, then suddenly it isn't
 

Hmm, despite the so-called "timing issue", there isn't a single post in their forum complaining about it.
 
So it begs the question:  Is this a perceived problem only from reading the specs, or have you guys actually demoed it to hear if it's unplayable in real-time?