afaiac you will always be best off with ‘close but not clipping’ samples that you can then dial back on the device than samples with too much headroom that you may need to boost … however, being pragmatic there may be a case for a modest -3db on all so there’s not constant dialling back … probably best to see what works out best in terms of workflow with the particular material you use … or search the internet for sound advice … this is just another opinion, but the quoted statement is wrong
while -6db gives a lot of headroom, there also is a requirement for legroom so to speak, so yes the file must have a proper representation as a waveform.
currently going through exporting one-shots from some original track element exports, essentially tweaked vocals and various synth/keys… and the 22050 rate actually sounds okay.
also exporting as monophonic as the MD doesn’t handle stereo samples, although i am not sure if the files need to be specifically exported as monophonic before import, to gain the memory filesize saving.
the 22050 sample rate exports do sound a bit muffled. and yet, i like them. and the 50% saving in filesize makes them even more of an attractive option.
i’m going through quite a lot of material, and so am unwilling to export each individual sample at both 44100 rate and 22050 rate.
if necessary i will return to the material and export 44100 versions of the more detailed sounds that would benefit from a higher hertz.
after some further exporting at 22050hz sample rate, i found some details and harmonics were lacking in drum samples, and so shall use 44100hz sample rate for drum one-shots, and maybe use 22050 for a longer chord or vocal, or a sample with a high-quality software-generated reverb tail.
actually i just have another quick question … if i prep samples at 24 bit (for further use in say the Digitakt, hypothetically speaking), will the Machinedrum accept them?
or do i need to just continue working at 16 bit and not have any format issues?
i understand the Machinedrum takes the resolution to 12 bits
although it is fascinating how a 22050hz sample rate is able to be preserved and thus allow for memory savings on longer files, i am currently working at 16 bit 44.1kHz
very cool, cheers
aha found the information … ‘all bit depths are accepted’… this is fantastico!
i like to future-proof my work session output, especially if hour upon hour is required to prepare 6000 samples to fill the Snapshot drive banks. 24 bit is much more current than 16 bit, and all my original projects are in 24 bit. It’s better for headroom so they say although i’m still confused exactly as to why.
interesting point to note as regards the kHz sampling rate… anything below 22050hz gets sampled up to 22050hz, anything above 44.1kHz will be sampled back to 44.1kHz
so it’s a two-tier system as regards memory usage and sampling rate. Although everything is going to be set to 12 bit resolution, it is possible to either use the 22050hz sampling rate or the 44100hz sampling rate.
Effectively resulting in a 50% memory filesize saving for each sample that sounds okay at 22050hz (samples with some noise or naturally less rich harmonics than other sounds).
the manual does not hand out this information, i found it when searching for info about the 24bit question.
of course, it is clear to me now after some proper thought that merely turning down the individual volume is going to ensure signal flow purity
i always turn down the volume of all drums to midday (vol dial pointing vertical), as that setting is the true “100 percent volume”, anything further is increasing the signal strength of the base waveform.
so if i was going to apply some additive eq, such as pushing 80hz on a bassdrum perhaps, i should ensure the individual volume is set back further than 12’o’clock, to about 10’o’clock.
i’m still going to apply a modest (very) reduction in volume during preparations so each clip has about 2.5db headroom below digital 0db clipping point.
If you want to save space, a good technique is to tailor sample rate to the particular file. A bass sound with very little harmonics will sound fine at a lower rate, a sound with high frequency content might need that higher rate to replicate properly.
Of course, if you are doing in bulk, this is probably not an option (although I bet there are some analysers out there that could detect minimally-useful sample rate).
As for processing in bulk, foobar2000 is really great for this - you can automate it and let it run on your entire collection hands-free.
lol i did try processing each sample individually one by one, both sample rates 22050 and 44100 but yes this is probably not an option. Lots of time and work. I wanted to try saving one export process for the sound files but the extra work (for eventually thousands of individual samples) was prohibitive. Formidable.
Then ealised that if i am exporting initially at 24bit 44.1kHz, to then later go over with a bulk exporter (cheers i shall check Foobar) exporting the 22.05kHz versions into a folder. Easy enough to A/B test each one, see how they sound.
22.050 is most assuredly okay for some sounds, although it is not always a desirable option.
In terms of overall sound, and sample choice for my previous Machinedrum,my best results were when i allowed it (him? her?) to be its true nature, that being a drumming drum machine.
also, on the subject of samples and what sounds cool… i found it a bit unpredictable… some guesswork and preparation could help predict if a sample would sound cool in the 12 bit MD environment, but the ultimate test was to put the file in there and trig the file, see how things sound.
Actually if i export a bank of samples at 22.050kHz i should also set them to be 12 bit, to help guess if they are going to be usable at that resolution too. Up until now i let the MD handle the 12 bit conversion.
Foobar2000 sounds like a top application for batching although Windows only…
For Mac, turns out that Audacity is able to do batch processing/exports…
One more question regarding volume levels… and signal path… Does the EQ section for each drum or sample get processed before or after the individual volume level?
I run foobar via wine on mac and it works very well.
I really wouldn’t recommend loading 24 bit samples into the machinedrum - it is my understanding that it doesn’t convert them to 12 bit on load, it just plays them out via 12-bit converters, so they might just take up extra space. I could be totally wrong about that and I’d appreciate if someone who knows would clarify.
Also have you considered 32kHz? If you really are talking thousands of samples (good luck importing all those to the MD), the space savings could be significant and the audio fidelity might not be an issue. I used to run a 32k sampler and I loved the sound of it.
32k - the new 12bit.
And I hope you are keeping the original samples regardless. And also that you know that upsampling or upconverting the bits will make no difference, so don’t do that to any of them.
many thanks for the post … yes lol 32kHz is the new 12 bit lolzor
oh dear, too funny.
cheers for the tips. yes am keeping the original samples as much of the chord keys and synth preps are original material to begin with.
thousands of sample preps, yes. load all into MD immediately? No way
the MD does actually perform the resolution conversion from 16 or 24 bit to 12 bit.
this process happens immediately upon import, before any usage of the sample takes place.
also, talking of auto-conversion, the information about 22050hz sample rate and 44100 sample rate doesn’t mention what happens to a 32000hz file.
the Machinedrum will auto convert anything below 22.05kHz,
to be 22.05kHz.
the Machinedrum will auto convert anything above 44.1kHz,
to be 44.1kHz
32kHz sampling rate is in between those two tiers of automatic conversion parameters.
Really would love to know if the EQ section for a ROM machine comes before or after the individual volume setting in the signal path flow.
The reason for this is that if EQ is being applied in an additive fashion, lowering the individual volume of the ROM machine will only help maintain unclipped signal if it comes before the EQ section.
The manual hands out what you search for. Give and it will give back.
All I did was opened the MD Manual PDF, searched for the terms “Bit depth”, and the search returned Page 71 as the one and only result.
oh that’s a lovely thing to say.
"
• All sample speeds from 4kHz to 48kHz are accepted and handled by the Machinedrum UserWave. Higher rates are accepted but the samples are downsampled to half the sampling speed."
but that is referring to what happens if a sampling rate is above 48000hz.
as relates to lower sampling rates of 22050 and 44100, all it says are that they are “accepted and handled”. Don’t get me wrong, I love the Machinedrum manual. Currently making an illuminated/illustrated version of it like a bohemian Book of Kells or something glitter glue for the win.
actually i was referring to the information about there being two internal tiers of resampling, 22.05kHz and 44.1kHz.
if a sample has a sampling rate of 5000 hz (lol, low quality but it is possible), it would seem the Machinedrum will not convert it to 44100hz but rather 22050hz instead.
and likewise, if a sample is 48000hz, the information i found stated that the Machinedrum would autoconvert it to 44100hz.
so i find that fascinating, that there are two tiers of sampling rate conversion, and it happens automatically within the Machinedrum according to certain parameters.
I am still not quite sure if it is true, but if it is, that means it is possible to theoretically put twice as many samples in the Machinedrum if targeting the 22050hz sampling rate ‘option’.
edit: or should i say, twice as much allowable sample time, within the allowable amount of samples (32 for mkI or 48 for mkII).
That is quite confusing. I wonder if there’s any way to test that…? Probably not… My guess is that it is as it says, and anything between 22–44khz is kept as-is.
As for downsampling to fit more in, I found that pitching up all of my samples an octave (thus effectively making 44khz audio 22khz when pitched back down) gave a good result. Coupled with a bit of SRR it gave a nice crisp sound. It’s a shame all (?) Elektron devices use interpolation on samples; I’d love to be able to get gritty without having to use SRR…
also found this page quite valuable …
