Performances generated using a reference performance and melody as conditioning signals.
| Conditioning Melody | Conditioning Performance | Generated Performance |
| Clair de Lune | ||
| Twinkle, Twinkle, Little Star |
Performances generated using a reference performance as a conditioning signal.
| Conditioning Performance | Generated Performance |
Interpolation in latent space between two performances originally from the MAESTRO dataset. Alpha denotes how much of the embedding from Performance A vs. Performance B is used (see paper for full details).
| Performance A (original) | |
|
| Alpha = 0 | |
|
| Alpha = 0.125 | |
|
| Alpha = 0.375 | |
|
| Alpha = 0.5 | |
|
| Alpha = 0.625 | |
|
| Alpha = 0.875 | |
|
| Alpha = 1.0 | |
|
| Performance B (original) | |
Interpolation in latent space between two performances originally from the internal dataset while keeping the melody conditioning signal constant. Alpha denotes how much of the embedding from Performance A vs. Performance B is used (see paper for full details).
| Original Melody (Clair de Lune) |
 |
|
| Alpha = 0 |  |
|
| Alpha = 0.125 |  |
|
| Alpha = 0.375 |  |
|
| Alpha = 0.5 |  |
|
| Alpha = 0.625 |  |
|
| Alpha = 0.875 |  |
|
| Alpha = 1.0 |  |