Method of generating music data

A computer-implemented method of providing one or more outputs at one or more respective time instants, the method comprising:

• generating at least one first data object executable to provide a first portion of an output, the at least one first data object comprising a parameter having a first value associated therewith;
• placing the at least one first data object in a first position in a sequence;
• generating at least one second data object executable to provide a second portion of the output, the at least one second data object comprising the parameter;
• generating a first array of probabilities by a first neural network for a second value of the parameter for the at least one second data object, the first array of probabilities being influenced by the first value, wherein the first neural network is trained to learn generation of sequences of notes in a musical way without consideration to an overall structure of a piece of music;
• generating a second array of probabilities by a second neural network for the second value of the parameter, the second array of probabilities comprising a probability that the second value is equal to the first value, wherein the second neural network is trained to learn generation of a musical structure indicating similarity between sections of the piece of music;
• combining the first array and the second array to provide a modified array of probabilities;
• determining and setting the second value based on the modified array of probabilities;
• placing the at least one second data object in a second position in the sequence, the second position providing a second portion of the output;
• and
• outputting the at least one first and second data objects at the respective first and second positions in the sequence to provide the output, wherein the at least one first and second data objects represent audio data or MIDI data.

A non-transitory computer-readable storage medium, storing executable instructions which are configured, when executed on one or more processors, cause the one or more processors to perform operations comprising:

• determining an initial sequence of notes for a piece of music;
• determining at least one probability distribution for selecting at least one subsequent note from a set of candidate notes, wherein the at least one probability distribution is generated by a melody generator, and wherein the melody generator is trained to learn generation of sequences of notes in a musical way without consideration to an overall structure of the piece of music;
• generating a structure probability array by a structure generator based on data of the initial sequence of notes, wherein the structure generator is trained to learn generation of a musical structure indicating similarity between sections of the piece of music;
• and
• extending the initial sequence of notes with the at least one subsequent note selected from the set of candidate notes based on the at least one probability distribution and the structure probability array, wherein the structure probability array biases a selection of the at least one subsequent note to affect a likelihood of the selection resulting in a repeat of a musical element formed by the initial sequence of notes.