Waivs-Ops: thousands of samples, loops & one-shots with CC-BY license

Related Projects
, ,
1

Hi @zynthianers!

Are you aware of this?

I downloaded the techno dataset and woaoooo!! Not bad at all! 12000 loops going from 128 to 150BPM. I copied the 128BPM dataset only (> 1000 loops) to my zynthian and i was “DJing” for hours. I simply created 4 x audio-clip columns, started loading loops and mixing.
I must say that the average quality is not bad at all. The only “fault” is not having any label apart from the BPM.

I tasted a little bit the HipHop and Trap datasets, and first impression is they are of similar quality. Perhaps not brilliant but not bad at all.

I’m thinking of creating some packages to install these data sets as “zynthian collections”.

Tell me if you test some of them.

Enjoy!

3 Likes
2

Throw in a couple of workflows to amuse and entertain us as you do it…

3

Hi @zynthianers and techno lovers!

I just published a set of sample collections based in the WaivOps EDM-TECH data set.

Captura desde 2026-07-18 16-34-29
Captura desde 2026-07-18 16-34-29994×1084 129 KB

They are 14 packages, each one containing about 1000 techno loops with the same BPM:

  • EDM-TECH-128BPM
  • EDM-TECH-130BPM
  • EDM-TECH-132BPM
  • EDM-TECH-134BPM
  • EDM-TECH-135BPM
  • EDM-TECH-136BPM
  • EDM-TECH-137BPM
  • EDM-TECH-138BPM
  • EDM-TECH-139BPM
  • EDM-TECH-140BPM
  • EDM-TECH-142BPM
  • EDM-TECH-145BPM
  • EDM-TECH-148BPM
  • EDM-TECH-150BPM

I converted all samples to flac format, what reduces the size to 1/4 of the original size.

From the WaivOps website:

EDM-TECH is an open-source corpus of drum recordings produced in the techno genre, consisting of 11,270 WAV files paired with JSON metadata for supervised model training. The dataset was generated using custom scripting applied to a proprietary database of MIDI patterns and one-shot drum samples. Recordings primarily feature drum machine and percussive drum synths, complemented by additional samples of synthesizers and chords. Data augmentation included random sample-swapping, track isolation, pitch-shifting, equalization, and convolution reverb modeling. These strategies enhance model generalization by exposing training examples to diverse rhythms, sonic qualities, and ambient effects. Training examples were randomly mixed with varying signal levels and audio qualities, providing examples that reflect the evolving styles of techno music over the decades.

The average quality of the samples has resulted to be better than i would expect for this kind of dataset. As a techno lover, i’m having a lot of fun DJing with this on Vangelis with my APC40 MIDI controllers, that is fully integrated, including AB-mixing and recently, PFL (pre-fader listening) by connecting an extra USB audio interface to my V5.

I’ve been DJing with other controllers too:

  • APC Key25
  • Launchpad Mini MK3 + MIDIMix

It’s quite comfortable too, but the APC40 is the best for DJing with Zynthian :wink:

I recommend you adjust zynthian BPM to match the data set BPM you are using, so no rewarping need to be done and loading samples is ultra-fast.

Of course, you can combine samples from all data sets, but then zynthian will rewarp the samples when loading, what takes a few seconds.

The sample names are not meaningful at all, they are untagged, so you don’t know what you are loading until you load it. Having tags would be very convenient. I’m thinking about it.

WaivOps have other datasets that i want to explore too. I simply started with the techno set because of my personal likes. If some of you want to step over any other data set, please tell me, so we can sync our efforts. I wrote some scripts to mass-process the datasets and generate the packages :wink:

Enjoy!

3 Likes