Thanks,
Yeah I watched Ricky’s video but he’s purposely destroying the sample.
I’m gonna resample the same sound multiple times without any tweaks inbetween.
Just because I’m an audio nerd and like to understand how stuff ticks.
4 Likes
Nice. Please report back with examples. I’m sure there are plenty of audio nerds (myself included) on here who’d be interested in the results 
+1 for posting nerdy Digitakt stuff 
I swear I’ve read someone posting audio examples where they resampled many times and noticed some very minor artefacts, then there was some discussion about that master EQ thing that I mentioned, I think there even was some official Elektron user commenting. It was a few years ago though.
But I can’t find the topic anymore 
Curious to learn the results. Best way is to compare the original with the processed and shifting the phase of one of them to hear the difference only.
OK fellow elektronauts, I’m back from nerding off with samples for a good hour or so while I could/should have been doing something productive 
.
My first experiment was a clean 440Hz sine wave, which I uploaded to the DT and the proceded to resample 10 times by iteratively assigning the new sample to the same track and resampling.
I didn’t find much difference between the 10th sample and the initial source, apart from an amplitude difference due to DT’s normalizing and a slight phase shift/glitch at the start of the sample, which is due to the trigger. This was expected, it’s a simple waveform after all.
So I won’t bother you with audio.
My second experiment was a more complexe waveform with various frequencies and transients, a realistic sample so to speak.
This time I resampled it 20 times.
It is noteworthy that the track had no FX at all, default values across the board.
The difference between sample_0 and sample_20 is already visually apparent (resampled waveform at the top):
Frequency analysis confirms this.
Sample_0
Sample_20
Obviously the proof is in the listening:
Sample_0
Sample_20
22 Likes
Nerd props!
I’m surprised how I instantly find the 2nd sample emotive, whilst the first was just ‘nice’. Probably need a psychologist to explain well why that is
1 Like
Well done !
So the spectral analysis shows there is marked amplification of higher frequencies, like from 5000 Hz on up, relative to frequencies below 5000 Hz.
Could this be a result of compression in this case with lower frequency parts being louder and being more reduced relatively by compression ?
Resampling white noise, and doing a spectral analysis on both might show this to be true or false.
2 Likes
Cheers for taking the time to share your experiment
The spectrum seems to imply a +1.x dB high shelf, I wonder if this is per track or applied on the master 
Were you resampling individual DT track or resampling the master?
1 Like
The track, not the master.
1 Like
That’s pretty interesting, It doesn’t really sound degraded either just EQ’d quite a lot.
I tried that with OT once, it will f*ck up your sample but 20 passes is nowhere near enough. It was very cool
I’ll just leave this here. I did some similar exploring in the past.
4 Likes
thanks for a super interesting experiment!
what in the world is this happening to that audio, and why? 
People complain about OT so they added degradation to make it sound better or something?
Seems to work, people like the tone
1 Like
I wonder if they have a similar thing going on in the Model:Samples or in the Rytm sample engine
Aliasing sounds great here. The pop in high freq content might be a part of internal engine, maybe filter. Would be cool to test it with filter disabled if it werent already.
Rytm sounded darker to my ears, which probably means it has no mid-top end boost like on DT
1 Like
Nice measurements! I too like to measure things.
10x seems to be enough to at least see a difference (-10dB?).
The test was done in a new Project without any parameter changes. The EQ was set to infinite hold.
#1: I uploaded a sine sweep (20Hz-20kHz@ -3dB).
#2: Uploaded into the Digitakt it seems to clip a bit. Kind of disappointing. Maybe due to the bit rate change from 48 to 16 though - wouldn’t read too much into it.
#3: After resampling the sample 10 times, a pretty specific EQ curve emerges.
Note:
I could have done more with a spectral graph, but I just wanted to share this. If you want more, let me know.
Maybe somebody could do a noise test, but the same EQ curve would probably be found.
3 Likes
Hey that looks pretty interesting! Very much in line of what @robbret found out.
I’m not really familiar with that monitoring tool (?), what are the blue bars and what are the yellow/red lines? I mean I get the gist of it – there’s a treble boost – but would be nice to be able to read the rest of these images.
1 Like
I used the Test Oscillator in Logic to create the sine sweep sample. I then monitored it with the Multimeter Analyzer.
You can ignore the blue bars. They represent the position of the sine wave sweep when I took my screen capture. The yellow lines are on infinite hold, so they are fully representative of the frequency response.
The yellow lines represent peaks. You can see #1 and #2 are relatively flat, but #3 shows a clear EQ curve as a result of the sampling.
The red lines represent clipping, but I wouldn’t read into it. It’s probably just from bit rate conversion in Logic which wouldn’t drastically change the frequency response.
I felt @robbret showed the time domain changes well, but I thought the frequency domain could be shown more clearly with a more stable measurement. I was also just curious what it would look like.
In a way (maybe more cutting than boosting). I tried to match this EQ curve. It looks like a 1.20Q @ 4kHz HPF. This curve would extend to just to around 8kHz like in the #3 example. However, there’s also a bit of high roll off and even more low roll off, so it’s definitely not just a simple HPF.
2 Likes