Description
Factoid is a Max For Live device by J.J. Burred that allows you to create endless rhythmical variations of audio clips. It uses machine learning to decompose a sound into up to 8 layers. Those layers can then be randomly shifted in time to change the rhythmical structure of the clip.
Factoid is not a loop slicing device: based on the same machine learning decomposition engine than Factorsynth (matrix factorization), it is often able to extract components that overlap in time in the original sample. For example, in a drum loop, it can often separate snare from kick even if they play at the same time in the original clip.
When a clip is loaded into Factoid, it is analyzed using matrix factorization. This produces a set of waveforms (the layers or elements) that are displayed superposed, each one with a different color. Each of the waveform layers can individually be shifted in time (left or right) by clicking on the “randomize” button. The interface allows to choose the number of desired components for factorization, and to control the time randomization in terms of quantization and frequency contents.
Factoid is a lightweight and easy-to-use device intended for live performance: you can sync its output with the Live set’s tempo, and adjust the clip’s pitch, just as with regular clips on session view.
lautopoiesis (verified owner) –
the idea is nice, but the spectral artefacts make the sound way too muddy..also the praised machine learning could be smarter, cause the shifted time layer don’t come up with musical useful results..
Darren E Cowley (Admin) –
Thanks for using Factoid, and sorry for the issues you’re experiencing. The artefacts are inherent to the decomposition into layers (they are leftovers from the other layers), and reducing them is an active research topic in matrix factorization. If we find a robust method to reduce those artefacts, it will be incorporated into future versions.
Also, please note that the machine learning part is only responsible for the decomposition into layers, it does not decide on the time shifts, which are completely random. In other words, there is no rhythmic modeling or generation in Factoid, just quantized randomisation.
Don’t hesitate to drop us a line if you need support!