Procedurally generating background music for sponsored audio

1. A computer-implemented method for generating audio content, the method comprising:

accessing sponsored audio information received from a third-party sponsor;

obtaining reference music features describing characteristics of reference songs;

determining music generation parameters based on the sponsored audio information and the reference music features;

producing, by a processor, procedurally-generated background music by inputting the determined music generation parameters to a procedural music algorithm that outputs music generated procedurally according to the music generation parameters;

generating sponsored audio based on the background music; and

providing the sponsored audio to a client device for presentation to a user.

2. The method of claim 1 , wherein the reference songs comprise feedback songs for which the user has provided feedback, wherein obtaining the reference music features comprises:

obtaining feedback features describing musicological characteristics of the feedback songs; and

determining the reference music features based at least in part on a combination of the obtained feedback features weighted according to a type of feedback given each feedback song.

3. The method of claim 1 , wherein the reference songs comprise a song presented previously by the client device, wherein obtaining the reference music features comprises:

obtaining the reference music features from music features of the previously presented song;

wherein the generated sponsored audio is provided to the client device for presentation in temporal proximity to the previously presented song.

4. The method of claim 1 , wherein the sponsored audio information includes a music generation constraint, wherein determining the music generation parameters comprises:

determining initial music generation parameters by applying a music generation model to the reference music features; and

determining the music generation parameters by modifying the initial music generation parameters according to the music generation constraint.

5. The method of claim 1 , wherein the music generation parameters include a category comprising a plurality of binary music parameters, wherein determining the music generation parameters comprises:

scoring each binary music parameter using a classifier that takes as input the reference music features; and

ranking the plurality of binary music parameters within the category according to the score of each binary music parameter; and

selecting a predetermined number of the binary music parameters according to the ranking.

6. The method of claim 1 , wherein determining the music generation parameters comprises determining a numerical music generation parameter based at least in part on a combination of the reference music features according to a set of weights included in a machine-learned music generation model.

7. The method of claim 1 , wherein obtaining reference music features describing characteristics of reference songs comprises obtaining acoustic features describing numerical characteristics of the reference songs.

8. The method of claim 1 , wherein the procedural music algorithm outputs the background music automatically according to rules that are configured by the music generation parameters.

9. The method of claim 1 , wherein the procedural music algorithm is at least partially dependent on randomness and outputs different background music from same music generation parameters.

10. A non-transitory computer-readable storage medium comprising computer program instructions executable by a processor, the instructions executable to perform operations comprising:

accessing sponsored audio information received from a third-party sponsor;

obtaining reference music features describing characteristics of reference songs;

determining music generation parameters based on the sponsored audio information the reference music features;

producing procedurally-generated background music by inputting the determined music generation parameters to a procedural music algorithm that outputs music generated procedurally according to the music generation parameters;

generating sponsored audio based on the generated background music; and

providing the sponsored audio to a client device for presentation to a user.

11. The storage medium of claim 10 , wherein the reference songs comprise feedback songs for which the user has provided feedback, wherein obtaining the reference music features comprises:

obtaining feedback features describing musicological characteristics of the feedback songs; and

determining the reference music features based at least in part on a combination of the obtained feedback features weighted according to a type of feedback given each feedback song.

12. The storage medium of claim 10 , wherein the reference songs comprise a song presented previously by the client device, wherein obtaining the reference music features comprises:

obtaining the reference music features from music features of the previously presented song;

wherein the generated sponsored audio is provided to the client device for presentation in temporal proximity to the previously presented song.

13. The storage medium of claim 10 , wherein the sponsored audio information includes a music generation constraint, wherein determining the music generation parameters comprises:

determining initial music generation parameters by applying a music generation model to the reference music features; and

determining the music generation parameters by modifying the initial music generation parameters according to the music generation constraint.

14. The storage medium of claim 10 , wherein the music generation parameters include a category comprising a plurality of binary music parameters, wherein determining the music generation parameters comprises:

scoring each binary music parameter using a classifier that takes as input the reference music features; and

ranking the plurality of binary music parameters within the category according to the score of each binary music parameter; and

selecting a predetermined number of the binary music parameters according to the ranking.

15. A system for generating audio content, comprising:

a processor; and

a non-transitory computer-readable storage medium comprising computer program instructions executable by the processor to perform operations comprising:

accessing sponsored audio information received from a third-party sponsor;

obtaining reference music features describing characteristics of reference songs;

determining music generation parameters based on the sponsored audio information the reference music features;

producing procedurally-generated background music by inputting the determined music generation parameters to a procedural music algorithm that outputs music generated procedurally according to the music generation parameters;

generating sponsored audio based on the generated background music; and

providing the sponsored audio to a client device for presentation to a user.

16. The system of claim 15 , wherein the reference songs comprise feedback songs for which the user has provided feedback, wherein obtaining the reference music features comprises:

obtaining feedback features describing musicological characteristics of the feedback songs; and

determining the reference music features based at least in part on a combination of the obtained feedback features weighted according to a type of feedback given each feedback song.

17. The system of claim 15 , wherein the reference songs comprise playlist songs having a playlist seed of a content playlist, wherein obtaining the reference music features comprises:

obtaining playlist features describing musicological characteristics of the playlist songs; and

determining the reference music features from the playlist music features based on frequency of each music feature among the playlist features;

wherein the generated sponsored audio is provided to the client device for presentation as part of the content playlist having the playlist seed.

18. The system of claim 15 , wherein the reference songs comprise a song presented previously by the client device, wherein obtaining the reference music features comprises:

obtaining the reference music features from music features of the previously presented song;

wherein the generated sponsored audio is provided to the client device for presentation in temporal proximity to the previously presented song.

19. The system of claim 15 , wherein the sponsored audio information includes a music generation constraint, wherein determining the music generation parameters comprises:

determining initial music generation parameters by applying a music generation model to the reference music features; and

determining the music generation parameters by modifying the initial music generation parameters according to the music generation constraint.

20. The system of claim 15 , wherein the music generation parameters include a category comprising a plurality of binary music parameters, wherein determining the music generation parameters comprises:

scoring each binary music parameter using a classifier that takes as input the reference music features; and

ranking the plurality of binary music parameters within the category according to the score of each binary music parameter; and

selecting a predetermined number of the binary music parameters according to the ranking.