Let me state clearly that I think Gigasampler is an amazing software sampler that has changed sampling for the better. On some bass slide tones that are not played at the original unity note some aliasing is present. It is a very minor problem however I think others in the GigaSampler community should be aware of this.
When working with GigaSamper on the New Orleans; The Delta Grooves sampling CD Rom I noticed that some aliasing was present on deep bass slide tones when the sample was triggered at a pitch other than the original unity note (in this case the tone was transposed up 1 semitone). You can hear the original bass tone in wav format http://www.numericalsound.com/bassorig.wav and the GigaStudio version transposed up one semitone (16 bit digital output) http://www.numericalsound.com/bassdgig.wav
Four graphs of the audio spectrums of these tones can be seen at the very bottom of the following web page the URL is http://www.numericalsound.com/smpcdgig.html The top two graphs are the audio spectrum of the original tone bassorig.wav in linear and exponential frequency representation . The bottom two graphs are the spectrums of the bassdgig.wav which contains the aliasing tones. Note that they are the spikes -70db down and in the frequency range of between 2000-11000Hz.
The work around solution for this sampling CD was to include a low pass bass filter in the GigaSampler patch. What I would like to see is an option for choosing a better interpolating algorithm - this might be at the expense of more voices but I think it is worth it. What do other users on this forum think ?
Thanks Ernest, I can hear what you\'re talking about - and that\'s a shame. It\'s that classic slightly \'metallic and airy\' fur around the signal.
You\'ve picked the kind of sound where filter performance is really highlighted, but then Giga is a \'pro\' piece of software for which many people are spending large sums of money on infrastructure, so I suppose they deserve better filter response, especially in an area as critical as bass.
BTW, the original wav file is about 6db \'louder\' than the giga version. This may confuse some people who download both and try and A/B them. Any chance of matching the levels of the two signals for those who are \'audio editor\' challenged?
Thanks again for pointing this out. I\'m sure it\'ll help Giga become a better platform.
Chadwick good point. The Giga file level has been increased by 6.33db so that both examples are at the same level.
KingIdiot I think that you are right most users would choose more voices (but I would definitely not). Whenever an instrument is featured in the foreground of an arrangement/mix however, then a higher quality interpolation algorithm would be important if one wants clarity in \"the mix\". THe software sampler Unity DS-1 has a option that allows one to select between 4 interpolation algorithms; none, linear, quadratic and Lagrange.
As a sampler developer I like clean bits ... and can settle for 16 clean as opposed to 24 dirty bits ! In this example look at the audio spectrum of the original bass it is approaching 96 dB. The interpolation process increases the harmonic distortion to .1% that may sound small but one can easily hear the difference in the examples I posted. True 16 bit is 0.003% distortion- something that I think we should shoot for considering that it is only a software issue.
On the sampling CD Drone Archeology the drone tones have 30-60 second loop lengths (all have a signal to noise ratio approaching 96db with 0.003% distortion). This way one can trigger the tone over several octaves without the familiar \"sound cycling effect\" (2 octaves up and it is still a 7+ second loop length). I could reduce the distortion to 16 bit levels by having a separate tone for each note however a 10meg file would turn into a 200+meg file. Thus a Giga interpolation selection option seems reasonable to me.
This is very interesting data. I\'d like to ask if you\'ve actualyl checked that different PCs create the same extra harmonics? This would be to eliminate the possibility that it is induced by a sound card or something else about the specific machine you did the work on.
The harmonics look interesting to me. Since the samples sound is a sliding bass note, wouldn\'t you expect a purely aliases signal to create a wide range of aliased harmonics? That\'s not what I\'m seeing visually.
What I see visually is that there is a root frequency at either 2.5KHz. (Or possibly 1.25KHz but I don\'t think so.) The I see harmonics off of that root at 5KHz, 7.5KHz and 10KHz. This makes me think that I\'m simply seeing and extra signal ADDED to your sample.
Clearly the low-pass would help if it helps supress the 2.5KHz peak, but that\'s not a great solution for a sample with more high frequency than this one.
It should not be a factor because it was sent out via the digital out from an EgoSys WT2496 to the same card\'s digital input on another computer.
I think you are correct it is not aliasing in the traditional digital audio sense, because the mirror image of this signal should be 44.1k-2k which is between 42.1-44.1KHz and instead it is in the 2-10k range. What this should be called is interpolation aliasing. The aliasing part of the sound can be heard without any of the original bass tone (2KHz High Pass Filter) in ound.com/bassdgia.wav\" target=\"_blank\">http://www.numericalsound.com/bassdgia.wav</a> you can hear some of the slide component in the original tone.
I created a looped sine tone at 50Hz and used the pitch bend on my keyboard controller to change the pitch. You can definitely hear the same type of aliasing sound even though the sample was triggered at its unity note. This means that any pitch transformation will results in interpolation aliasing whether it can be heard depends on the masking properties of the tone nevertheless it is always there.
I definitely do not think that GigaSampler software has a built in \"huge mistake\". The interpolation algorithms were probably originally designed several years ago. A reasonable tradeoff was made with processor speed versus maximizing the voice count. However, the pentium processor speeds have greatly increasing since then and with the broad digital audio push to 24 bits -I think that this area needs further software refinement so that this part of the system can also be SOTA.
In the next day or so I will post a giga file with a 50Hz looped tone so you hear this on your own sound card.
First of all, I\'m only so-so with FFT stuff, and it\'s been years anyway. Second, you\'ve given this a lot more thought than I, so I understand you\'re much further down the curve towards an explanation.
That said, I do have a lot of engineering experience and understand that often what appear obvious in the end isn\'t, and that things I often thought could not effect the outcome often do.
And if I keep my mind open and learn, these things can always be explained.
So, that said, my problem with this discussion so far is that it seems to assume that the original designers of GigaSampler made some huge oversight and missed this whole issue. While that may indeed be true, I\'m not ready (personally) to jump there just yet without some more study. (AND I\'m NOT saying you are, as you have thought and studied this more than I!!)
Clearly if the mirror image signal did exist, it might very well just be filtered by other parts of the system as there is probably some sort of filter to roll of everything above 20KHz anyway.
I\'m not where I can listen really critically to the samples right now to hear the effects you\'re talking about in the latest post. I\'ll have to revisit that this evening.
And please do not take me wrong. I think this work is very interesting and more of this sort of stuff needs to be done. I suppose in life I\'m paid to be a doubting Thomas, so it carries over everywhere. I still think this needs to be checked on another PC. If you could put the 50Hz loop gig file online somewhere for others like me to check out, that would indeed be helpful and I think lend itself to more thoughtful discussion.
If there is a file I could capture usign my PC and send to you for spectrum analysis, I\'d be happy to do that.
As promised there is a GigaSampler patch at <a href=\"http://www.numericalsound.com/giga50hz.zipttp://www.numericalsound.com/giga50hz.zipe. This tone has a slow frequency modulation and a high pass filter to reduce the 50Hz tone so that one can easily hear the aliasing in the treble part of the sound (you will hear a \"drizzle\" sound).rical Sound