A project by Vie McCoy
tags: tensorflow, machine learning, audio signal processing
break.Gen is a machine learning project that remixes any song of any genre into a Drum and Bass / BreakCore remix with varying levels of musical success.
The program first splits the stems from the selected track. Then, it analyzes the drum patterns. Depending on the complexity and arrangement of the drums, break.Gen intelligently replaces the percussive elements from the track with BreakCore samples
BreakBeat replacement of the original percussion is done through the use of an algorithmic energy score, which is calculated via the amount of drum hits and total amplitude.
Users can specify the desired BPM, intensity (which controls the amount of samples chopped in each bar), and whether or not they would like hyperpop-style formant shifted vocals.
The above settings produced the following with zero additional post-processing:
Hotel California - break.Gen remix