DestrucTango AI

By Juan Ramos & Pablo Riera

This page was made for DestrucTango AI, our proposal for an Artistic Mini Project for Music
RAI
. It is not listed on this website.

Here we provide 5 demos made with our 2ASMRS AI model (https://github.com/pabloriera/2ASMRS). The two featured datasets (orchestral and bandoneon solo) used to train the model were created either from our own material, or recordings currently under public domain.


Demo 1

Acoustic bandoneon performance, used as partial input for a bandoneon-dataset trained 2ASMRS.

La Cumparsita as rendered by 2ASMRS trained with a bandoneon dataset (performance by Joaquín Sterman).

Original performance of La Cumparsita as played by Joaquín Sterman.


Demo 2

A performance with “Alfa“, our Electronic Bandoneon controller, used as partial input for a bandoneon-dataset trained 2ASMRS.

Mano a Mano as rendered by 2ASMRS trained with a bandoneon dataset (performance by Joaquín Sterman).

Original performance of Mano a Mano as played by Joaquín Sterman on “Alfa“, our Electronic Bandoneon.


Demo 3

An orchestral version of A Media Luz, as reinterpreted by 2ASMRS trained with an orchestral Tango dataset. The model produced interesting percussion-like sounds not present in the original recording, but still part of the Tango expressiveness. It also seems to have preferred only minor mode chords, rather than the major+minor keys from the original.

A Media Luz, as reinterpreted by 2ASMRS trained with an orchestral Tango dataset.

Original version of A Media Luz, used as partial input for 2ASMRS.


Demo 4

A re-orchestrated version of Mano a Mano by Carlos Gardel. We took the guitar stems as input for 2ASMRS and mixed its output over the original recording, keeping Gardel’s voice and a bit of guitar to aid in the perception of the harmonies. In the future, we will explore DestrucTangoing the voice stems too.

Mano a Mano, with guitar parts reinterpreted by 2ASMRS trained with an orchestral Tango dataset.

Original version of Mano a Mano by Carlos Gardel.


Demo 5

An orchestral version of Corrientes y Esmeralda, as reinterpreted by 2ASMRS trained with an orchestral Tango dataset. In this case the model produced more extreme sounds but retained most rhytmic features from the original.

Corrientes y Esmeralda, as reinterpreted by 2ASMRS trained with an orchestral Tango dataset.

Original version of Corrientes y Esmeralda, used as partial input for 2ASMRS.