related-systems · wu2024-realchords

Some of these systems are asynchronous, where the user first provides the full melody, and the system generates an accompaniment offline. Examples include MySong (Simon et al., 2008), where a user sings a melody and the system generates chords to accompany them. Most recently, SingSong (Donahue et al., 2023) supports a very similar interaction, but generates full-band backing tracks. Both are offline systems. In contrast, online accompaniment systems need to synchronize with user actions in real-time. Score-following is a special case where the system has the score [...]. Music Plus One (Raphael, 2010) adapts its playback speed of an orchestral recording (without the soloist) to a soloist's expressive performance. Similarly, Antescofo (Cont, 2008) follows where a soloist is in a score and triggers live electronics accordingly. Generative accompaniment systems or more generally co-creative music systems, not only have to anticipate user actions, they need to learn how to respond. Voyager (Lewis, 2003) takes a rule-based approach [...], while Omax Brothers (Assayag et al., 2006) recombines what a musician plays on-the-fly as an accompaniment [...]. ImproteK and later DJazz [...] leverages a shared predefined chord progressions (such as a Jazz Standard) to coordinate the human-machine improvisation. [...] BachDuet (Benetatos et al., 2020) trains an LSTM model using MLE for counterpoint (melody to bassline) accompaniment. SongDriver (Wang et al., 2022) focuses on online melody-to-chord accompaniment