ChordRipple: Adaptively Recommending and Propagating Chord Changes for Songwriters

Abstract

Songwriting is the interplay of a composer's creative intent and an idiom's language. This language both facilitates and poses stylistic constraints on a composer's expressivity. Novice composers often find it difficult to go beyond common chord progressions, to find the chords that realize their intentions. To make it easier for composers to experiment with radical chord choices and to prototype "what-if" ideas, we are building a creativity support tool, ChordRipple, which (1) makes chord recommendations that aim to be both diverse and appropriate to the current context, (2) infers a composer's intention to help her more quickly prototype ideas. Composers can use it to help select the next chord, to replace sequences of chords in an internally consist manner, or to edit one part of a sequence and see the whole sequence change in that direction. To make such recommendations, we adapt neural-network models such as Word2Vec to the music domain as Chord2Vec. This model learns chord embeddings from a corpus of chord sequences, placing chords nearby when they are used in similar contexts. The learned embeddings support creative substitutions between chords, and also exhibit topological properties that correspond to musical structure. For example, the major and minor chords are both arranged in the latent space in shapes corresponding to the circle-of-fifths. To support the dynamic nature of the creative process, we propose to infer a composer's intentions for adaptive recommendation. As a composer makes chord changes, she is moving in the embedding space. We can infer a composer's intention from the gradient of her edits' trace and use this gradient to help her fine-tune her current changes or to project the sequence into the future to give recommendations on how the sequence could look like if more edits in that direction were performed.