Research prototype
Record once.
Double it naturally.
Doubletracker is an experimental neural guitar doubler trained on real DI pairs. The goal is to generate a second performance of the same part, creating a stereo image that feels closer to real doubletracking than algorithmic doubling or widening.
The current model is still early, and reconstruction quality remains a work in progress.
01 / LISTEN
Current model examples
Compare the recorded references, the real doubletrack, and the model output. For the stereo mixes, use headphones or properly placed monitors.
Prototype note: All examples are currently limited to a 16 kHz sample rate. Recorded reference tracks pass through the same bandwidth-limited processing stage for a like-for-like comparison. Higher sample rates are planned as the model develops.
02 / CURRENT STATE
Still in development.
Doubletracker is currently a research prototype. The model already produces convincing doubletracking on some material, but quality and generalization are still being improved.
03 / CONTRIBUTE
Have double-tracked DI?
The most useful contribution is real, independently recorded doubletracks that you own—or have permission to provide for model training.